In all experiments, we used a miniSIA-2 instrument (GlobalFIA, USA). Unless otherwise specified, this instrument featured two High Flow milliGAT pumps with stainless steel heads, a plexiglass 8-port Lab-On-Valve (LOV) affixed to a selector valve (VICI, USA), two thermostated holding coils set at 25°C, and a 20 cm Linear Light Path (LPP) flow cell mounted to the LOV side ( Figure 1 ). A Tungsten Halogen light source and a VIS-NIR Flame Spectrophotometer (OceanInsight, USA) were connected to the flow cell using 500 µm silica core optical fibers sheltered in 0.8 mm I.D PEEK tubing (GlobalFIA, USA). A 40 psi flow restrictor (Idex Health & Science, USA) was placed at the flow cell outlet to minimize degassing in the light path. Except for the 1.6 mm I.D tubing between the carrier reservoirs and milliGAT pumps, the LOV ports tubing connections and holding coils (1,300 µL) were all 0.8 mm I.D polytetrafluoroethylene (PTFE) and were furnished with ¼-28 fluidic connectors (Idex Health & Science, USA). The reagents, standard solutions and ultra-pure water carrier were stored in sterile, flexible storage bags ranging from 50 mL to 3 L (Flexboy, Sartorius, USA), hung adjacent to the instrument, and connected to the COV ports using ¼-28 fluidic connectors (Idex Health & Science, USA).

During the International Nutrient Intercomparison Cruise (INIV), we integrated an external autosampler (GlobalFIA, USA) to the miniSIA-2 for efficient analysis of full depth vertical profiles. This autosampler featured a 14-port selection valve, linked to 14 acid-washed 30 mL screw caps polypropylene vials (Sarstedt, Germany). One end of the LOV flow-through sample port connected directly to the central port of the autosampler selection valve, with the opposite end of the LOV sample port linked to the autosampler’s diaphragm pump ( Figure 1 ). This arrangement ensured sample integrity by preventing the sample from making direct contact with the diaphragm pump. In all experiments, the entire setup, including the miniSIA-2 components and the autosampler, was controlled by FloZF software version 5.2 (GlobalFIA), operating on a Dell Latitude 3590 laptop running Windows 10.

All experiments employed plasticware acid-cleaned for at least 24 hours in a 10% HCl bath (v:v). We made all working reagents and stock standard solutions using high-purity milliQ (MQ) water (> 17 MΩ cm^-1). Primary standard salts were dried for 24 hours at 100 °C, then allowed to cool to room temperature in a desiccator before weighing with 0.1 mg precision. Stock standards (1 L) of phosphate (0.7 mM) and silicate (4.97 mM) were prepared by dissolving KH₂PO₄ (0.0954 g) and Na₂SIF₆ (0.4718 g, purity 99.4%) in MQ water in class A volumetric flasks. All volumetric flasks and pipettes used for standard preparation were calibrated gravimetrically Becker et al. (2020), and the resulting stock standard solutions were stored refrigerated in opaque High Density Polyethylene (HDPE) bottles.

The working standards, depending on the application, were created by serial dilution of the concentrated stocks in either MQ water or low nutrient seawater (LNSW) collected from the Hawaii Ocean Time-series station ALOHA (22°45 N, 158°00 W). The LNSW was filtered at sea through a 0.45 µm Acropak capsule filter (Pall, USA) and stored in acid-cleaned HDPE cubitainers in the dark at room temperature until use. To prolong stability during unattended deployments (> 5 days), the PO₄ working standards should be acidified 13.3 mM using 6 M HCl (reagent grade).

The spectrophotometric determination of phosphate and silicate is based on the reaction of the analytes with molybdate in an acidic medium, followed by the reduction of the resulting molybdic complexes to phosphomolybdenum blue (PMoB) and silicomolybdenum blue (SiMoB) using ascorbic acid as a reductant (Murphy and Riley, 1962; Truesdale and Smith, 1976). The silicate assay includes an extra step with the addition of oxalic acid to prevent the reduction of excess molybdates and, presumably, the decomposition of interfering phosphomolybdic species (Thomsen et al., 1983). The assays measure dissolved orthophosphate and orthosilicic acid. However, they may also detect contributions from colloidal materials and, in the case of phosphate, labile organic phosphates that are reactive to the assay conditions.

In a pFI assay protocol, the analyzer ( Figure 1 ), governed by software scripts, mixes the sample and reagents at a user-defined ratio. The ratio is orchestrated by the two pumps, which are connected to the Lab-On-Valve’s central port via the holding coils (HC) and operate synchronously but in opposite directions during mixing. Typically, one pump aspirates the sample into HC1 while the other dispenses reagent previously held in HC2 to HC1. The aspirating pump sets the ratio. For example, if 200 µL of sample is aspirated while 100 µL of reagent is dispensed, a 1:1 ratio is established. The difference between the aspirated and dispensed amounts represents the reagent volume added to HC1 (in this case 100 µL), with the remainder (100 µL) as the sample. A portion of this mixture is then transferred to the flow cell, where it remains until the chemical reaction reaches steady state and the resulting absorbance is recorded ( Figure 2B ). This step is achieved by positioning the center of the reacting mixture in the flow cell, ensuring that its boundaries are outside the light path, thereby eliminating Schlieren effects (Hatta et al., 2021).

We developed a new pFI protocol for phosphate analysis ( Figure 2A , Supplementary Table S1 ). First, 600 µL of sample is aspirated into HC1. The sample subsequently mixes with molybdate in a 4:1 ratio in HC2, forming the phosphomolybdic complex. This complex then merges with ascorbic acid, also at a 4:1 ratio. A 400 µL segment of this reacting mixture is then delivered to the flow cell, where the flow is stopped for 60 seconds and the absorbance of the PMoB product is measured. Finally, the system is flushed with 2 mL of MQ water from both pumps, preparing it for the next sample ( Supplementary Table S1 ). Our phosphate protocol provides distinct benefits. Aspirating the sample first allows for online dilution of a concentrated MQ standard, facilitating autonomous system calibration. Moreover, this protocol is advantageous for long-term unattended operations, consuming only 80 µL of molybdate and another 80 µL of ascorbic acid per sample.

For silicate analysis, we adopted the pFI protocol outlined in Hatta et al. (2021), which is detailed in Supplementary Table S2 . One key modification was adjusting the holding coils’ temperatures to 25°C instead of the originally specified 50°C. Our decision to use a lower temperature stemmed from concerns regarding equipment wear and potential safety issues when maintaining a holding coil unattended at 50 °C for extended durations. In tandem with this temperature change, the stop flow time in the flow cell was extended to 60 seconds to ensure the absorbance of the SiMoB product reached a steady state ( Figure 2C ).

The VIS-NIR Flame spectrophotometer was configured with a 30 ms integration time. Using the FloZF software, the resulting counts were smoothed by averaging six consecutive spectral scans and four consecutive pixels. To correct for potential baseline offsets, absorbance values obtained when the flow cell was filled with MQ water were subtracted from reacted sample absorbances.

For phosphate analysis, the absorbance of the PMoB product was computed using the following equation: ABS_PMoB = (ABS₈₈₀ – ABS_880baseline) - (ABS₁₀₅₀-ABS_1050baseline). Employing a reference wavelength at 1050 nm, as opposed to solely relying on a single measurement wavelength at 880 nm, enhanced the precision of the measurements. To accommodate the analysis of Si samples from full depth ocean profiles (up to 180 µmol L^-1) within the constraints of a 20 cm flow cell, a dual measurement wavelength approach was adopted. For low Si samples (< 40 µmol L^-1), the difference between the baseline corrected absorbances at 810 nm and the 1050 nm reference was used. For high Si samples (> 40 µmol L^-1), a singular, baseline corrected measurement wavelength at 660 nm was employed. This strategy extends the linear range of measurements using the same 20 cm optical path used for phosphate analysis, whilst preserving sensitivity at the lowest Si concentrations. As a result, the same instrument can be used for sequential analysis of phosphate and silicate in all but the most oligotrophic regions of the ocean. Phosphate and silicate concentrations in the samples were calculated using the slope and intercept from calibrations in either MQ or LNSW, contingent on the application (specified below).

We investigated how molybdate concentration and acidity in the reaction mixture influence analytical sensitivity and salt effects of both pFI assays. For the silicate assay, acidities ranging from 1.5 to 30 mM [H⁺] were tested at molybdate concentrations of 0.5, 0.75 and 1 mM. For phosphate, we tested acidities from 0 to 1 mM at molybdate levels of 0.5, 1.0 and 1.5 mM. Analytical sensitivity was estimated from the slope of 3-point MQ calibration curves for each acidity-molybdate combination. To evaluate salt effects under each condition, LNSW was spiked with known concentrations of the analytes and measured using the corresponding MQ water calibrations. Salt effects were defined as the percentage error between the measured and known spiked LNSW concentrations, with errors greater than 5% indicative of suboptimal conditions for seawater analysis.

To evaluate the long-term stability of our optimized reagents, we tested each reagent solution under a range of temperatures representative of seasonal fluctuations at Moss Landing, CA, our first unattended pFI deployment location. We first generated 3 point-calibration curves for phosphate (0-2 µmol L^-1 PO₄) and silicate (0-30 µmol L^-1 Si) using freshly prepared reagents and standards. The solutions, stored in darkened flexible bags immediately after preparation, were then placed in a water bath at three different temperatures for 36 to over 90 days, depending on the treatment. Temperatures were continuously recorded using a HOBO pendant datalogger during each treatment (Onset Corp, USA). At set times, we connected the aged reagent and standard solutions to the miniSIA-2 pFI and repeated the calibrations. For the phosphate assay, we repeated the calibrations using aged solutions every 3-5 days for 40-50 days at 7, 13 and 24°C. For silicate, the calibrations were only performed initially and after 36-92 days of incubation time at 7, 10 and 24°C.

We defined stability based on the change in calibration slope at time t relative to the initial slope obtained with freshly prepared solutions (t₀). If significant degradation occurred, indicated by a >10% change in slope, fresh reagents were prepared in succession and new calibrations were run until the initial slope was re-established. This systematic reagent replacement approach allowed us to identify which reagent or standard had degraded and provided insights into the long-term integrity of each solution under simulated deployment temperature conditions.

We evaluated the shipboard capabilities of the pFI analyzer during two research cruises. In the North Pacific Ocean aboard the RV Kilo Moana in September 2022, we collected unfiltered duplicate samples from a full-depth profile (n=36, 5-4800 m, cruise # HOTS-339). One set was analyzed in triplicate for phosphate using the pFI analyzer calibrated with MQ standards (0-3 µmol L^-1). The duplicate samples were frozen, shipped to Scripps Institution of Oceanography (SIO) Ocean Data Facility chemistry laboratory on dry ice and analyzed at SIO’s lab using an AA3 autoanalyzer (Seal Analytical, USA). The resulting pFI and SIO vertical profiles were compared using a Type II linear regression.

During the International Nutrient Intercomparison Cruise (INIV) in the Southern Ocean aboard the RV Investigator in June 2023, the pFI analyzer analyzed seven full-depth profiles for silicate along the SR03 transect and one profile at the Southern Ocean Time Series (SOTS) site. For maximum accuracy, we calibrated the pFI using six working standards prepared in LNSW, and these standards contained both Si (0-119 µmol L^-1) and PO₄ (0-2.8 µmol L^-1). Daily quality control included analyzing a low nutrient seawater blank (LNS30, OSIL, UK) and three Reference Materials for Nutrients in Seawater (RMNS, Kanso Technos Co, Japan) at the beginning of each run. This was followed by the analysis of 12 unfiltered seawater samples. In the autosampler sequence, these samples were arranged with two RMNS bracketing them for quality insurance. This process was repeated three times each day to analyze all 36 samples. We used freshly opened RMNS bottles each day but kept the same bottle of LNS30 throughout the 13-day voyage.

In both silicate and phosphate assays, the concentration of molybdate and acidity of the reaction mixture greatly influences analytical sensitivity and the magnitude of blank signals. Previous research has extensively covered these effects for phosphate (Nagul et al., 2015; Ruzicka et al., 2019), but similar investigations for silicate are limited. Additionally, the role of molybdate and acidity on salt effects has been underexplored in both assays.

Our data reveal a narrow optimal range of acidity for both assays that optimized sensitivity and reduced salt effects across all molybdate concentrations tested ( Figure 3 ). For silicate, doubling the molybdate from 0.5 to 1 mM improved sensitivity by ~30% but this came at the expense of precision, as illustrated by the larger confidence intervals of the slopes at 1 mM molybdate. Furthermore, we observed that the input molybdate reagent for silicate precipitates within a few days of storage at concentrations exceeding 4 mM (> 0.5 mM in flow cell after mixing, Figure 3 ), regardless of the paired acid concentration.

Given our aim to develop a versatile Si method for diverse applications within the constraints of a 20 cm flow cell, we concluded that 0.5 mM molybdate and 12 mM H⁺ provided the optimal reaction conditions, striking a balance between sensitivity, input reagent stability and matrix effects ( Figure 3 ). Yet, for benchtop use where the molybdate reagent can be freshly prepared daily, increasing the molybdate concentration in the Si reagent to 0.75 mM would also be appropriate ( Figure 3 ), particularly for samples < 30 µmol L^-1- Si. Interestingly and unlike phosphate, we observed that replacing H₂SO₄ with HCl in the molybdate reagent for silicate amplified the salt error (data not shown).

For the phosphate assay, our results align with Ruzicka et al. (2019) in terms of sensitivity and blank effects when HCl is the source of acidity. However, we newly identify 0.5 mM and 100-150 mM acidity as ideal for reducing salt effects ( Figure 3 ). Notably, any increment in the molybdate concentration beyond 0.5 mM does not enhance sensitivity, instead it merely exacerbates salt errors ( Figure 3 ). In addition, we explored the potential interference from silicate in the phosphate assay, which is particularly sensitive to the concentration of H⁺ in the reacting mixture (Nagul et al., 2015). We compared the absorbance readings of MQ water standards spiked with 180 µmol L^-1 Si to a MQ blank, and the absorbance of a 3.0 µmol L^-1 MQ phosphate standard with and without the addition of 180 µmol L^-1 Si. The maximum absorbance difference between the treatments containing 3.0 µmol L^-1 phosphate with and without 180 µmol L^-1 Si added was 0.002 AU at 880 nm ( Supplementary Figure S1 ). This difference corresponds to an apparent phosphate concentration of 0.015 µmol L^-1. This value falls within the precision of the assay (4%) and thus does not compromise the accuracy of our phosphate determinations.

Our results further challenge the exclusive use of H⁺/Mo ratios as an optimization metric for both assays (Nagul et al., 2015; Ruzicka et al., 2019), since various combinations of molybdate and H⁺ yield the same ratio, yet some combinations are less suitable for seawater analysis. Our optimization work also underscores the previously overlooked role of molybdate and acidity on matrix effects — only a narrow range of acidities result in minimal salt effects. This observation may explain conflicting reports of salt effects in existing methods relying on the molybdenum blue assays for silicate and phosphate analysis (Hansen and Koroleff, 1999; Grand et al., 2017; Deng et al., 2020; Hatta et al., 2021; Fang et al., 2022a), although variations in sample to reagent ratios and analytical configurations undoubtedly contribute.

The optimization work outlined previously identified the most suitable molybdate-acid pairs that minimized salt effects — an important consideration for the analysis of samples of varying salinity. However, this screening process — using LNSW samples with a known analyte concentration against a MQ calibration curve — does not confirm that a MQ calibration curve can be applied to seawater samples. Instead, the decision requires careful inspection of the slopes and intercepts of calibration curves obtained using standards prepared in MQ and LNSW, as depicted in Figure 4 .

For phosphate, calibration slopes obtained in MQ and LNSW can be identical, indicating consistent PMoB formation rates in both matrices. Yet, a marked difference was observed in the intercepts, with the MQ calibration from this example showing a significantly lower value ( Figure 4A ). This difference of 6.1 mAU, termed the matrix blank (Hatta et al., 2021, 2023), will introduce a positive offset of ~0.03 µmol L^-1-PO₄ when applying the MQ calibration to seawater samples. For silicate, the 95% confidence intervals of the slopes and intercepts in MQ and LNSW overlapped at 880-1050 nm (< 40 µmol L^-1- Si). At 660 nm, the calibration intercept in MQ water was significantly lower than in LNSW ( Figure 4A ), suggesting that a LNSW calibration is preferable when analyzing high Si concentration samples at 660 nm (>40 µmol L^-1- Si). Thus, the decision to apply a MQ water calibration for seawater samples hinges on the accuracy requirements of the application and anticipated sample concentrations. While errors stemming from the application of MQ calibrations to seawater samples may be corrected (see Hatta et al. (2021, 2023)), this adjustment assumes constant slopes and intercept ratios between MQ and LNSW calibrations. The long-term validity of this assumption remains to be tested.

The long-term stability of reagents and standards is a critical consideration for long-term autonomous deployments. Although a range of wet chemical analyzers suitable for long-term deployments are commercially available, literature on reagent longevity is limited. To the best of our knowledge, Li et al. (2019) is the only study that evaluated reagent preservation methods for the colorimetric reagents of the phosphomolybdenum blue assay. Their findings, however, may have limited applicability since their input reagents were markedly more concentrated than ours. In addition, the stability of the silicate reagents, including the oxalic acid, has never been reported.

The ascorbic acid reagent is the most susceptible to degradation, since it can be oxidized by light, heat, transition metal ions contaminants and alkaline conditions (Yin et al., 2022). For instruments containing acrylic manifolds (e.g., Grand et al. (2017) and this work), a surfactant must be included in the ascorbic acid reagent to maintain the linearity and sensitivity of the phosphate assay. This requirement presents an additional challenge when input reagents are exposed to ambient temperatures during unattended deployments, since the optimal surfactant for the pFI phosphate assay, Sodium Dodecyl Sulfate (SDS), precipitates at high concentrations and temperatures below 15°C (Adornato et al., 2007; Clinton-Bailey et al., 2017). For this reason, we lowered the SDS concentration in the ascorbic acid input reagent from the 170 mM originally used by Hatta et al. (2021, 2023) to 8.7 mM. This modification effectively prevented SDS precipitation over two months of cold storage in a refrigerator (5-10°C), without compromising analytical performance (data not shown).

After adjusting the SDS concentration in the ascorbic acid reagent, we proceeded to evaluate the stability of the colorimetric reagents across a temperature range of 7-24°C. At all temperatures tested, the calibration slopes obtained using ascorbic acid/SDS reagents incubated for up to 52 days remained within 10% of the initial values ( Figure 5A ). We note that aged ascorbic acid reagent solutions exhibited signs of oxidation, evident from a yellow coloration that induced a noticeable elevation in absorbance spectra near 400-500 nm. However, since this waveband does not overlap with any of the wavelengths used in both assays (660, 880, 810 and 1050 nm), analytical performance was not compromised. Thus, our experiments suggest that the complex ascorbic acid preservation methods outlined by Li et al. (2019) are unnecessary when the ascorbic acid is combined with SDS and shielded from light in a sealed storage bag within 7-25°C.

Both molybdate solutions, when stored as specified for up to 42 days, maintained analytical sensitivity within 10% of initial values. In fact, in a separate experiment, our optimized formulation for the input molybdate reagent for Si — characterized by a reduced Mo concentration (< 4 mM) to prevent precipitation and complete loss of sensitivity—remained within our 10% threshold at 10°C for up to 92 days and so did the oxalic acid ( Supplementary Figure S2 ). However, because our experiments were primarily focused on analytical sensitivity, we did not evaluate if the aging of oxalic acid affects its ability to prevent the reduction of excess molybdates and/or decomposition of interfering phosphomolybdic complexes.

While the colorimetric reagents met our stability threshold for at least 42 days, the absorbance of 1.0 µmol L^-1 phosphate standards prepared in MQ water declined by more than 10% within 5-10 days ( Figure 5B ). This behavior, which indicates analyte loss from the solution, likely arises from the adsorption of phosphorus anions on the walls of plastic and glass storage bottles (Hassenteufel et al., 1963; Zhu and Ma, 2020). In our study, utilizing flexible storage bags made of ethylene vinyl acetate (EVA), we observed this adsorption loss to occur at a rate of approximately 20 nmol L^-1 phosphate per day. An established strategy to reduce these adsorption losses involves acidifying the standards to a pH of 2 (Grand et al., 2017). The addition of acid shifts phosphorus speciation towards the fully protonated phosphoric acid (H₃PO₄), reducing the proportion of charged, absorbable anions—specifically H₂PO₄ ⁻in the working standards. Figure 5B confirms that acidification to pH 2 effectively minimized adsorption losses at all temperatures tested for up to 35 days. Using acidified standards (13.3 mM H⁺) to calibrate the instrument does not significantly impact the accuracy of phosphate determinations in unacidified samples ( Supplementary Figure S3 ). This is also supported by the constant level of sensitivity of the phosphate assay at H⁺ concentrations ranging from 150 to 200 mM in the presence of 0.5 mM molybdate ( Figure 3 ).

In contrast to phosphate, the Si assay’s sensitivity to H⁺ concentration precludes the use of acidified Si standards. Specifically, any increase in acidity beyond the optimized concentration of 11.2 mM H⁺ will lead to a significant decrease in analytical sensitivity and greater than 5% errors when applying a MQ calibration curve for seawater sample analysis ( Figure 3 ). However, since the predominant Si species at pH 5.6 (=MQ pH) is uncharged (H₄SiO₄), silicate standards prepared in MQ water should suffer fewer adsorption losses than phosphate.

In summary, our reagent stability experiments demonstrate the durability of the colorimetric reagents in both pFI assays, including the ascorbic acid, which does not require specific treatment to prolong its reduction efficiency. The longevity of the reagents meets the needs for unattended pFI deployments at mid-latitude shore-stations or underway shipboard applications with monthly servicing intervals. Acidifying the working standards to pH 2 is a practical strategy to minimize adsorption losses for phosphate. However, more work is needed to confirm the stability of unacidified silicate standards.

Having established the durability of the reagents and the analytical performance of our newly optimized pFI assays under controlled laboratory conditions with human oversight, we performed a series of pilot autonomous deployments to demonstrate the capabilities of pFI for autonomous analysis. These short-term deployments, lasting less than a week, were conducted at a coastal shore station, a commercial seaweed aquaculture facility and aboard a research vessel as a “FerryBox” type system. Our intention is not to provide a rigorous evaluation of the long-term capabilities of pFI analyzers. Rather, we aim to showcase the versatility and unique benefits of pFI analyzers for autonomous nutrient analysis, setting the stage for future, more ambitious autonomous deployments and comprehensive performance evaluations.