The NP Sensor architecture is based on the orthophosphate sensor system described in our previous work (Morgan et al., 2022), with differences noted here. All alterations to the previous system that were needed to realize the combined dual-species instrument will be highlighted as such. Sensor electronics were distributed across 3 custom printed-circuit-boards (PCBs), responsible for automation, data processing and logging, and controlling the various sub-instruments needed to propagate fluids. A total of three stepper motors were used to actuate four syringes, each used to propagate the various types of fluid used during sensor activity. Our syringes were custom-built to better withstand wide temperature fluctuations that may occur during a deployment and were driven using a custom syringe-pump assembly. A separate, pressure compensated “solenoid can” housed a total of 10 solenoid valves (LFN series, Lee Company, Connecticut, United States) which were operated using a “spike-and-hold” approach to minimize power consumption when active. The solenoid valves were used to route fluid through the various fluid paths and vias machined into our custom lab-on-chip (LOC) platform for dual-species analysis. Fluid channels had a square cross-section, with nominal dimensions of 400 µm by 400 µm. In Figure 1A, the LOC is shown sandwiched between the main sensor body (below) and the solenoid can (above).

The fluid schematic of the dual-chemistry instrument is shown in Figure 2. Fluids may enter or exit the sensor through one of six fluid ports. In Figure 2A, these ports are labelled NR, T, S, PR1, PR2, and W, and are named according to their function during the NP protocol. Ports NR, PR1, and PR2 were used to draw in the modified Griess reagent for the nitrate assay (nitrate reagent–NR), and the two phosphate reagents required for the phosphomolybdenum blue assay to quantify phosphate. Similarly, ports T and S were used to draw in either calibration standards of known nitrate and phosphate concentration, or sample water during deployments. Finally, the waste port W was used to eject consumed fluids from the sensor into a waste reservoir. Each port except the sample intake S was connected to their respective fluid reservoir via 1/8 tubing (fluorinated ethylene propylene, FEP, Cole-Parmer Canada). The sample intake, however, was designed to draw directly from the surrounding environment. A disposable 0.45 µm syringe filter (Sarstedt Inc., PN 83.1826) was attached to the sample intake via a luer-lock connection for use during field deployments. This pore size was used to prevent clogging of microfluidic channels and valve pathways from large particulates, while also eliminating optical losses from scattering when analyzing highly-turbid samples (Hansen and Koroleff, 1999). The sample intake filter can be installed or replaced underwater if desired but could lead to small amounts of unfiltered water entering the fluid pathways. All filter installations were performed on land as a precaution.

Four syringes were responsible for fluid propagation throughout the sensor and are arbitrarily labelled in Figure 2A for reference. S1 was used to push nitrate reagent through the short path measurement cell and was controlled using the first of three stepper motors. Similarly, S3 and S4 were used to push phosphate reagent 1 and phosphate reagent 2, respectively, through the long path measurement cell and were both controlled by the second stepper motor. Finally, S2 was used to push either sample or standard fluid through both short and long path measurement cells and was driven by the third stepper motor.

Two parallel inlaid optical cells were used to obtain measurements. Figure 2B shows a photograph of the top layer of the bilayer lab-on-chip. Both inlaid optical cells are highlighted by dashed-stroked rectangles. For each optical cell, light emissions from one of two light sources within the sensor are directed towards the top 45°C prism. A portion of the light reflected off the prism is directed through the corresponding fluid channel (directly downward from the prism in the photograph), where the fluid channel crosses the clear-black material, towards the bottom prism. Light transmitted through the fluid sample is then measured using a light detector. Light sources and detectors are fixed to a PCB in the sensor mounted directly above the chip.

Nitrate measurements used the “short path” optical cell of path length 10.4 mm to interrogate a fluid volume of just under 2 µl. A green light-emitting diode (LED) centered at λ₁ = 527 nm (Cree C503B-GAN-CB0F0791-ND, FWHM ≈ 50 nm, Durham, NC, United States) was directed through the short path optical cell for measurement using a photodiode (TSL257, AMS, Premstaetten, Austria). Similarly, the “long path” optical cell was used to measure phosphate using an infrared LED centered at λ₂ = 880 nm (MTE8800NK2, FWHM ≈ 60 nm, Marktech Optoelectronics, New York, United States) and a second photodiode. Two reference photodiodes also were used to monitor and subtract light fluctuations in the LEDs.

A detailed description of the sensor protocols used to perform each series of experiments can be found in the Supplementary Material. A custom-built scripting language was used to operate basic sensor functions, including actuation of motors, actuation of solenoid valves, LED intensities, fluid volume tracking, data processing, and data logging. A typical protocol to acquire nutrient measurements of a fluid sample followed the following sequence of steps. First, logging of engineering-related metrics is enabled, which includes statistics regarding power draw, internal temperature, and valve-related parameters. Next both short and long path LEDs are assigned their own output intensity levels ranging from 0.0–100.0. LED intensities are typically determined prior to field deployments in a controlled setting to obtain a wide dynamic measurement range through the respective measurement cell without saturating the photodiode. Next, fluid propagation of the various reagents/samples/standards occurs according to valve and pump commands. Finally, after pumping, photodiode measurements are recorded during stopped flow for a set period of time (depending on the type of measurement) to acquire the necessary measurement data to produce concentration measurements.

Unless otherwise stated, chemicals were of analytical grade and supplied by Fisher Chemical (Waltham, MA, United States). Benchtop optics calibrations with stable fluids used red food dye and copper (II) sulfate standards that ranged from 0.00125%–0.05% and 0.625 mM–25 mM, respectively. Five concentrations of red food dye were prepared to characterize the performance of the short path measurement cell. First, a 0.05% (v/v) red food dye stock solution was prepared by diluting 0.5 ml of red food colouring (commercial food colouring, Club House Canada) with Milli-Q to a total volume of 1 L and mixing thoroughly. Four more standards were prepared through serial dilutions to produce a total of five standards of the following red food dye concentrations: 0.05%, 0.025%, 0.0125%, 0.0025%, and 0.00125%. Similarly, five concentrations of copper (II) sulfate were prepared to characterize the performance of the long path measurement cell. First, a 25.0 mM stock of copper (II) sulfate was made by adding 3.1217 g of copper (II) sulfate pentahydrate ( C u S O 4 ⋅ 5 H 2 O , CAS-no. 7758–99–8, Sigma Aldrich, Missouri, United States) to approximately 200 ml of Milli-Q water in a 500 ml volumetric flask. The flask was then filled to volume with Milli-Q, capped, and mixed thoroughly. Four standards were then made through serial dilutions to produce a total of five calibration standards of the following concentrations: 25 mM, 12.5 mM, 6.25 mM, 1.25 mM, and 0.625 mM.

Simultaneous nitrate and phosphate calibrations used mixed standards containing both nitrate and dissolved orthophosphate, otherwise referred to as “NP standards”. Six combined standards were prepared to evaluate the full dual-species protocol. To prepare the dual-nutrient standards, a stock solution containing 1,000 µM P O 4 3 − was first prepared by dissolving 0.1361 g of potassium phosphate monobasic ( K H 2 P O 4 , CAS 7778–77–0) in Milli-Q to a volume of 1 L. The first NP standard was then prepared by mixing 10 ml of the 1,000 µM phosphate stock with 14 ml of a 442.7 ppm nitrate nitrogen standard (Cat. No. 5457–16) and diluting to a final volume of 1 L with Milli-Q. The concentration of the first NP standard was 100 µM nitrate and 10 µM phosphate. The remaining NP standards were prepared through serial dilutions to realize six standards of the following concentrations: 100 μM and 10 μM, 50 μM and 5 μM, 25 μM and 2.5 µM, 10 and 1 μM, 5 μM and 0.5 µM, and 2.5 μM and 0.25 µM (listed as [ N O 3 − ] & [ P O 4 3 − ]).

Finally, the preparation of the nitrate reagent, and of both phosphate reagents, used for each colourimetric assay followed previous works (Murphy et al., 2021; Morgan et al., 2022). Nitrate Reagent was prepared such that 2.5000 g of vanadium (III) chloride, 15 ml of concentrated hydrochloric acid (HCl), 0.1250 g of NEDD (N-(1-Naphthyl)ethylenediamine dihydrochloride, CAS-No. 1465–25–4, Sigma Aldrich) and 1.2500 g of sulfanilamide (CAS-No. 63–74–1, Fisher) was mixed with Milli-Q to a total volume of 500 ml. Similarly, Phosphate Reagent 1 was prepared such that 0.5600 g of ammonium molybdate (VI) tetrahydrate ( H 24 M o 7 N 6 O 24 ⋅ 4 H 2 O , CAS-no. 12054–85–2) was mixed with 6.73 ml of concentrated sulfuric acid ( H 2 S O 4 , CAS-no. 7664–93–9, EMD Millipore, Darmstadt, Germany), 12.5 ml of a 4.5 mM stock antimony potassium tartrate solution ( C 8 H 10 K 2 O 15 S b 2 ⋅ 3 H 2 O diluted with Milli-Q, CAS-no. 28300–74–5, Sigma Aldrich, Missouri, United States), and Milli-Q to a final volume of 1 L. Lastly, Phosphate Reagent 2 combined 10 g of L-ascorbic acid ( C 6 H 8 O 6 , CAS-no. 50–81–7, Sigma Aldrich, Missouri, United States) and 0.10 g of polyvinylpyrrolidone (PVP) ( C 6 H 9 N O n , CAS-no. 9003–39–8, 10,000 g/mol, Sigma Aldrich, Missouri, United States) with Milli-Q to a final volume of 1 L. All reagents were refrigerated near 4°C between use.

Figure 3 shows the results from a benchtop calibration using stable fluids. Standards of red food dye (RFD) and copper (II) sulfate ( C u S O 4 ) were injected into either short or long path optical cells from highest to lowest concentration. For simplicity, these will be referred to as “dyes” due to their stable color. Milli-Q was used to flush both optical cells and to record baseline absorbance measurements between each set of dye measurements. In Figures 3A,B, Milli-Q flushes are marked by unshaded regions. Black diamonds indicate photodiode data during pumping while blue squares indicate “blank” measurements obtained from Milli-Q during stopped flow. In the shaded regions, dye measurements are observed as red circles in Figures 3A,B, respectively. A sharp drop in voltage data is observed when either dye is first injected into their optical cell. A steady plateau is then observed, when pumping is finished, from which absorbance measurements are calculated. The consistent photodiode measurements during each period of stopped flow demonstrates the optical stability of our system over both measurement cells.

Figures 3C,D plot absorbance vs. dye concentration for each fluid. Small error bars are shown, computed as the standard deviation of each triplicate absorbance measurement, indicating a strong precision between repeated measurements. Error bars in Figures 3C,D are small in magnitude with a maximum relative error of 4.8%. In the short path data, the largest error observed was 4.8 mAU on the 25 mM standard, but most standard errors were less than 1 mAU. In the long path data, the largest error observed was 0.8 mAU. A linear fit was applied to both datasets showing high correlation (R ² > 0.999). The root-mean-square error of the red food dye and C u S O 4 calibration was 10.6 mAU and 1.6 mAU, respectively–these correspond to inaccuracies of 0.0008% RFD and 54 µM C u S O 4 . The residuals of each calibration are shown in Figures 3E,F for RFD and C u S O 4 , respectively.

The molar attenuation coefficient for red food dye and copper (II) sulfate can be calculated by dividing the slope of their absorbance vs. concentration calibration curve by the optical path length. The short and long path optical cells had lengths of 1.04 cm and 2.54 cm, respectively. Using the convention σ R F D = V d y e / V s o l u t i o n , the attenuation coefficient of each species is calculated such that ϵ R F D = 1220.6   ( σ R F D ⋅ c m − 1 for red food dye and ϵ C u S O 4 = 0.0114   µ M ⋅ c m − 1 for C u S O 4 . The results of the combined sensor are in good agreement with previous results using single-species sensors over similar concentration ranges: ϵ R F D = 1100   ±   50   σ R F D ⋅ c m − 1 (Luy et al., 2020) and ϵ C u S O 4 = 0.0120   ±   0.0014   µ M ⋅ c m − 1 (Morgan et al., 2022). The ∼10% larger attenuation coefficient for red food dye observed here is reasonable and could be attributed to differences in protocol (manual vs. automated), or variability in LED emission spectra or food colouring batches.

Based on the data in Figure 3, the combined optical and fluidic stability of the system can be assessed and used to calculate theoretical limits of detection (LODs) for each optical cell. The average photodiode noise observed throughout 15 independent blank measurements was 0.5 mV ± 0.3 mV for the short path and 0.47 mV ± 0.14 mV for the long path. Using three times the photodiode noise [according to the triple-sigma literature method for LOD analysis (Shrivastava and Gupta, 2011)], and a typical blank baseline value of 2 V to convert to absorbance units, an LOD of 0.3 mAU and 0.31 mAU is calculated for both short and long paths and is consistent with previously-reported findings (Morgan et al., 2022). Using the correlation slopes provided in Figures 3C,D, the theoretical LOD of 0.00003% RFD and 11 µM C u S O 4 is found for our combined dual optical cell system. Overall, these results demonstrate the expected linear relationships between light absorbance and analyte concentration and absorbance with strong reproducibility, precision, and accuracy.

After dye calibration, a dual chemistry nitrate and phosphate calibration was performed. A series of six calibration standards, each containing varying amounts of nitrate and phosphate, were analyzed from most-to-least concentrated. Figure 4 shows the results from a benchtop nitrate and phosphate calibration at room temperature. Figure 4A shows the raw short path photodiode voltage data during the test, and was used to measure the nitrate concentration of the standard. Figure 4B shows the raw long path voltage data during the same test, and was used to measure the phosphate concentration of the standard. In Figures 4A,B, the unshaded and shaded regions denote blank and sample measurements, respectively. Blue (square) data points correspond to photodiode measurements during stopped flow after flushing with blank. Red (circle) data points correspond to stopped-flow photodiode measurements after pumping standard and reagent into the optical cell. In Figure 4A, standard and nitrate reagent combine to produce the mixed sample in the short optical path. In Figure 4B, standard and both phosphate reagents combine to produce the mixed sample in the long optical path. The raw voltage time-series data begins with the first sample triplicate measurement with its prior blank measured before the test began (data not shown). The usual pattern is then observed whereby blank triplicates precede each sample triplicate.

Figure 4C plots short path absorbance vs. the final nitrate concentration in the measurement cell after mixing with reagent. The mixing ratio between standard and reagent was 1:1; therefore, the reported concentrations of Figure 4C are half those of the prepared standards. A linear regression was applied, and a relationship between short path absorbance and nitrate concentration of A = 0.0053 [ N O 3 − ] + 0.0208 was determined. A strong degree of linearity (R ² = 0.998) was observed with a root-mean-square error of 4.3 mAU. Using the slope of the calibration curve, this corresponds to a root-mean-square error on sample measurements of 0.80 µM N O 3 − (1.6 µM feed concentration). Figure 4D plots long path absorbance vs. the final phosphate concentration in the measurement cell after mixing with both reagents. Absorbance values are calculated using a development time of 10 min to be consistent with our previous works. The mixing ratio between standard and both reagents was 2:1:1 and thus plotted concentrations are half those of each standard. A linear regression was applied such that A = 0.0428 [ P O 4 3 − ] with strong linearity (R ² = 0.999). A root-mean-square error between the regression fit and the data of 2.9 mAU was calculated: this corresponds to a feed concentration error of 0.14 µM.

Residual analysis of both nitrate and phosphate regression fits are shown in Figures 4E,F, respectively. The magnitude of the largest residual was 8.6 mAU (nitrate) and 4.1 mAU (phosphate); these occurred for nitrate and phosphate standard concentrations of 50 μM and 5 µM (feed concentrations). Therefore, over the tested range of nitrate and phosphate standard concentrations, a maximum error in accuracy of 3.2 µM and 0.19 µM is observed at room temperature conditions for either measurement cell. It should be noted that while the accuracy of our nitrate detection is ∼1/20 that of phosphate, the concentration range is 10 times larger. Similarly, the optical path length is 10.4 mm while that of the phosphate cell is 25.4 mm. This trade-off between accuracy and range was chosen to accommodate typical deployment conditions whereby nitrate concentrations are typically higher than phosphate.

These results can be used to determine the limit of detection (LOD) and limit of quantification (LOQ) of our sensor for both nutrient assays, where the limit of quantification instead uses 10x the blank baseline noise and is considered a more practical metric for sensor performance (Shrivastava and Gupta, 2011). The nitrate LOD and LOQ results are as follows. An average error of 0.4 mV was measured over 15 repeated blank measurements; using a blank baseline of 2 V, this corresponds to an LOD of 0.26 mAU. Using the slope of the calibration curve shown in Figure 4C, 0.0053 AU/µM, a feed concentration LOD of 97 nM and LOQ of 324 nM is determined for our instrument at room temperature conditions. Similarly, the phosphate LOD and LOQ results are as follows. An average error of 0.5 mV was measured over 15 repeated blank measurements; using a conservative but typical blank baseline of 2 V, this corresponds to an LOD of 0.3 mAU. Using the slope of the calibration curve shown in Figure 4D, 0.0428 AU/µM, a feed concentration LOD of 15 nM and LOQ of 49 nM is determined. These limits for phosphate detection compare well with the results of our previous study at room temperature: LOD = 15.2 nM and LOQ = 50.8 nM (Morgan et al., 2022).

The properties of both color development assays must be characterized at a variety of temperatures before deployment. Using the dual chemistry protocol, three calibrations were performed, each at a fixed temperature: T = 5°C, 10°C, and 15°C. The results of the temperature characterization of our system are shown in Figures 5, 6. In Figure 5A, short path absorbance vs. sample nitrate concentration is plotted for each of the three temperature calibrations. A linear regression with forced-zero intercept is applied to each data set with parameters outlined in Table 1. An exponential relationship between measurement sensitivity S and temperature T was then determined from each calibration slope: S N O 3 − = 0.0001956 e 0.1445 T , where S is the measurement sensitivity in units of AU/µM and T is reaction temperature in °C. This exponential relationship is shown in Figure 5B as a dashed line. Figure 5C shows the agreement between measured short path absorbance values and the applied linear fit, with residual errors computed for each temperature data set. The magnitude of the largest residual was 1.2 mAU, observed in the 10°C test, and corresponds to a feed concentration error of 2.8 µM. Similarly, a residual analysis of the fit applied to the nitrate sensitivity vs. temperature data is shown in Figure 5D. The magnitude of the largest residual was 4.8 × 10 − 3 mAU/µM and is approximately 3 orders of magnitude lower than the measured slope values. The residual errors shown in Figure 5D have a small relative influence vs. those of Figure 5C and are ignored when considering sources of error.

Figure 6 shows the phosphate results from the same dual chemistry temperature calibration study. Figure 6A plots long path absorbance vs. sample phosphate concentration for each temperature calibration. A linear regression with forced-zero intercept is applied to each data set with parameters outlined in Table 1. For each temperature, the most concentrated sample was omitted from the applied fit to optimize sensor performance for “typical” environment conditions (i.e. [ P O 4 3 − ] < 5 µM). The outlier behavior of these highest-concentration samples is attributed to a minor concentration-dependent reaction rate of the PMB assay at colder temperatures ((Clinton-Bailey et al., 2017), supplemental material), while at room temperature the reaction kinetics converge to a consistent rate (Sjösten and Blomqvist, 1997). These findings are consistent with our own, where the results of our phosphate calibration at room temperature in Figure 4C were linear up to the same upper concentration limit. Furthermore, when considering the data in Figure 6A, the agreement of the highest concentration sample to the applied trend improves from coldest to warmest temperature. An RMSE of 1.5 mAU was found when omitting the highlighted samples from the analysis: a significant improvement on the results shown in our previous work for a similar temperature sweep and range of standard concentrations.

In summary of the phosphate results, when measuring samples of feed phosphate concentration 5 µM or lower, a relationship between sensitivity vs. temperature of S P O 4 3 − = 0.00090 T + 0.025 should be used to minimize error from data fitting, as shown in Figure 6B (dashed line). Consequently, the residual errors obtained from these optimized fitting conditions, shown in Figures 6C,D, show a significant improvement on our previously published results.

In August 2022, a 1-week shallow water deployment was performed at a junction between Sackville River and the northern tip of the Bedford Basin (44°43′45.0″N, 63°39′44.4″W). A sensor package containing the NP Sensor, battery pack, and fluids was deployed with an RBR Brevio³ CTD and a Xeos iridium beacon (Xeos Technologies Inc., Dartmouth Canada). A large, perforated cage was used to contain the various instruments without causing water stagnation and sat on the bottom of the river. The depth of the NP Sensor ranged from 0.3 m–2.0 m depending on tides and precipitation, while the water temperature ranged from 17°C–24°C.

Figure 7 combines sensor data measured in situ throughout the Sackville River deployment. In Figures 7A,B, nutrient measurements obtained from the NP Sensor are shown with accompanying temperature and depth measurements from the CTD. For each nutrient measurement, the ratio of measured nitrate to dissolved orthophosphate was computed and plotted in Figure 7C. Water level and discharge measurements obtained from a sensor located approximately 400 m upstream from our deployment location are overlaid to contextualize N:P fluxes [obtained from the Government of Canada, Station 01EJ0011, 44° 43′ 53″ N 63° 39′ 37″ W, (Government of Canada, 2022)]. Finally, Figure 7D plots the measured nitrate and phosphate concentration of an on-board standard. The on-board standard was analyzed periodically throughout the course of the deployment as a method of in situ performance assessment.

In Figure 7A, the measured nitrate concentration at the start of the deployment was high (roughly 10 µM) and dropped over the course of the first two days towards near-zero. Over the next few days, an out-of-phase pattern between nitrate concentration and water depth can be observed, where the lowest sensor measurements typically occurred at high tide (and highest measurements at low tide). Finally, over the last three days, nitrate levels are observed to fluctuate more drastically in comparison to the previous 5 days. Similarly, in Figure 7B, the measured phosphate concentration over the first two days began at roughly 1.5 µM and dropped on average until the mid-point of the second day. The measured phosphate levels remained relatively consistent throughout the course of the first four days compared to the fluctuations observed in the nitrate results. However, a sharp increase in measured phosphate is observed beginning on the 24th. Phosphate concentrations spiked up to roughly 3 µM and remained high from the 25th until the end of the deployment.

A Fast Fourier Transform (FFT) analysis of the phosphate measurement data revealed a frequency peak corresponding to a period of 12.7 h: consistent with the corresponding 12.3 h peak observed in the tidal data captured by the CTD within a 3.2% difference. The phase difference between phosphate and tidal measurements appears to be roughly 180°C (out-of-phase). On the other hand, FFT analysis of the nitrate measurement data did not show this tidal frequency peak. We speculate that phosphate concentrations at our deployment location measured out-of-phase with tides due to dilution from water sourced by the Bedford Basin which was of relatively low phosphate concentration. This argument is supported by external data obtained from a DOT Phosphate Sensor deployed off the Centre for Ocean Ventures and Entrepreneurship’s Stella Maris Multi-Sensor Seabed Platform (MSSP) during the same period. Deployed at a depth that ranged between 8 m–10 m, the sensor measured phosphate concentrations below 1 µM in the Bedford Basin throughout our 8-day deployment in the river. A study by Shi and Wallace in 2018 demonstrated two-layer flow in the Sackville River, where the top layer is comprised of freshwater discharge by the river towards the basin, while the bottom layer is comprised of higher salinity water sourced by the basin (Shi and Wallace, 2018). Our sensor was stationed on the bottom of the river and therefore was most influenced by dilution from the basin.

The measured concentrations of both nitrate and phosphate are used in Figure 7C to calculate the measured nitrate-to-phosphate ratio (N:P ratio) over time. A spike in the N:P ratio is observed beginning midday on August 23rd and corresponded with a significant rainfall event that persisted throughout the day. At the same time, water level and discharge data obtained from the Government of Canada at Station 01EJ0011 (400 m upstream) indicates a corresponding spike in water level and discharge from the Sackville River. As the water level and discharge data returns to the previous levels reported the day prior, the measured N:P ratio also returns to the pre-spike levels. It should also be noted that the N:P ratio observed during the first 1.5 days of the deployment was also high on average. While not shown, a similar spike in water level and discharge occurred a few days prior on August 17th. From these findings, we speculate that nitrate concentrations in the river are most influenced not by tidal behavior, but instead by loading through runoff and/or sediment disruption.

A total of 18 individual standard measurements were made throughout the 8-day deployment: these results are plotted in Figure 7D. Standard measurements were performed in triplicate and occurred at 6 intervals each (approximately) 30 h apart. The concentration of the on-board standard was [ N O 3 − ] = 25 µM and [ P O 4 3 − ] = 2.5 µM. The average nitrate concentration reported by the NP Sensor over the 18 standard measurements was 24 µM ± 3 µM. This absolute uncertainty, computed as the standard deviation of all measurements, corresponds to a relative uncertainty of 13%. Similarly, the average phosphate concentration reported by the NP Sensor over the 18 standard measurements was 2.0 µM ± 0.3 µM (a relative uncertainty of 16%). These results show an excellent agreement between the measured and actual nitrate concentration of the standard within the reported error bounds. On the other hand, while the reported phosphate concentration falls slightly below the true concentration of the standard (80%), this finding is consistent with the under-reporting of the 2.5 µM phosphate standard described during the benchtop testing in Section 3.3 during temperature testing. Overall, the NP Sensor showed consistent measurement of the combined nutrient standard throughout the course of the deployment in a real environment of varying temperature conditions.