We carefully followed procedures as described in the literature to ensure methodology was as close as possible to that originally described (Tables S1 and S2). Three different voltammetric setups were used (Table S1). A standard Metrohm set up was used for TAC. For use with SA, a separate Metrohm system was modified to allow for air purging whilst the mercury drop formation was still executed under nitrogen pressure. Nitrogen did not leak into the headspace of the sample during the measurements in our Metrohm stand. However, when drops are formed pulses of nitrogen are released and end up in the headspace of the sample, and purging with air would remove (at least part of) the nitrogen. To check a potential effect of this nitrogen, a kinetic experiment with SA25 was executed. To five identical subsamples SA was added at the same time. These subsamples were each measured repetitively during 1 h to several hours, one after the other. No effect was seen after subsample replacements (Fig. S1). We concluded that nitrogen from the stand did not influence the kinetic process, since the measured FeAL concentrations had a gradual change over time, independent of the subsample. For the other kinetic experiments with SA, BASi equipment was used (Table S1). The electrochemical settings used by Croot and Johansson (2000), Buck et al. (2007), and Abualhaija and van den Berg (2014) were used without alteration and are summarized in Table S2.

Electro-active complexes with Fe typically have low solubility and thus tend to adsorb on the walls of containers. Conditioning and equilibration of all contact surfaces is thus an important pre-treatment step in order to minimize losses of Fe and ligand species during the course of the experiment. Different materials do not have the same adsorption properties (Fischer et al., 2006). All cells and titration vials were made of Teflon. Other bottles, sample bottles and those used for kinetic experiments were low-density polyethylene (LDPE) bottles (Nalgene™, Fisher Scientific). For all three applications, the same materials were used, canceling any deviation among methods from the interaction of solution component and containers. Equilibration between the samples and AL was attained at room temperature.

Before use, all materials such as vials, bottles and cells were conditioned overnight in UV-irradiated seawater with the prepared combinations of seawater and ligand. The conditioning procedure was performed at least three times for the analysis with TAC and at least five times for analysis with SA. The cell with electrodes, stirrer and purge tube were kept overnight in low-metal seawater. Before a titration was started, first two measurements were executed with seawater containing all chemicals but no Fe addition. These measurements also served as check for possible contamination of the cell. Hereafter, two zero additions were measured (see Sect. 3.4), of which the second was used as the start of the titration.

Before starting kinetic measurements, a 30 mL vial with the same content and treatment as the sample was used for three analyses (thus three times 10 mL) in order to condition the cell wall, electrodes and stirrer.

The 200 mL bottles used for kinetic studies were conditioned with 6 nM Fe, in the absence of the added ligand. For tests with UV-irradiated seawater without a model ligand, 200 mL bottles were conditioned by rinsing the bottle three times for 2 min with 30 mL of the test seawater. Since UV-irradiated seawater did not contain Fe-binding organic ligands, most of the added 6 nM DFe would adsorb on the bottle walls or precipitate.

Samples were equilibrated according to the specific method descriptions, which was overnight equilibration for the TAC and 5 µM SA and 15 min for 25 µM SA (Croot and Johansson, 2000; Buck et al., 2007; Abualhaija and van den Berg, 2014). The 15 min equilibrations were applied precisely using a stopwatch, whereas overnight equilibration resulted in a period of at least 14 h.

Samples, without any additions, were poured into 30 mL Nalgene FEP bottles and placed in a custom-made UV box between 4 TUV 15W/G15 T8 fluorescent tubes (Phillips) for 4 h (Rapp et al., 2017; Wuttig et al., 2019). Precipitates were not observed. After UV irradiation, samples were transferred into a clean 1 L trace-metal LDPE bottle and kept in the refrigerator.

The following discrete synthetic ligands of known concentration (model A ligands) were used at a concentration of 2 nM, unless otherwise stated. No tests were undertaken to check the purity of the siderophores. The solutions were used within 2 weeks after preparation and kept in the refrigerator in the dark at 4 ∘C, which should at least for DFOB be short enough to prevent degradation (Hayes et al., 1994). The aim of our research was to compare the three applications. Humics (model B ligands, 0.1 or 0.2 mg/L) were added in a concentration to give an iron-binding capacity of approximately 3 nM (Laglera and van den Berg, 2009; Yang et al., 2017; Sukekava et al., 2018). The stoichiometry of the formed Fe–model ligand complexes differs for each model ligand. In order to simplify the comparison of binding strengths, stability constants are given for a 1:1 stoichiometry.

Diethylenetriaminepentaacetic acid (DTPA C14H23N3O10, Sigma-Aldrich D6518-5G) was used to calibrate the added ligands via reverse titration according to methods described previously (Croot and Johansson, 2000). We calculated a conditional binding constant logKFeDTPA,Fe3+cond of 19.0 using the ion-pairing speciation software Visual MINTEQ (Gustafsson, 2012), disregarding the formation of FeOHDTPA. The logKFeDTPA,Fe3+cond value was independent of the pH difference 8.05–8.2. This is 0.34 higher than the value (18.65) used by Croot and Johansson (2000) at I=0.7, pH = 8.05, with the difference most likely arising as a result of the lower ionic strength predicted by the ion-pairing model. Addition of 2 and 4 nM DTPA did not increase the DFe content of the UV-irradiated seawater (detection limit = 25 pM)

Humic substances are the heterogeneous mix of hydrophobic compounds originating from chemical and microbial transformation of living matter as it decays in the environment.

Fulvic acid (FA) is the smaller and more soluble fraction of humic substances (Buffle, 1990). Therefore, this is not just a ligand but also a series of compounds of which a fraction function as iron-binding ligands. Laglera and van den Berg (2009) determined that 1 mg of this specific FA binds 16.7 ± 2.0 nM Fe with KFeL,Fe'cond=10.6. (IHSS Suwannee River Fulvic Acid Standard II, 1R101F). Yang et al. (2017) and Sukekava et al. (2018) found that 1 mg could bind 14.6 ± 0.7 nM Fe. (IHSS Suwannee River Fulvic Acid Standard II 2S101F). It must be noted that the batches are different between the above results, and these values can differ per produced batch. Still we assumed 2.92 nM equivalents (nM Eq) of ligand sites to be added with 0.2 mg SRFA per liter.

Fifteen vials were prepared with increasing Fe content, in a mixture of UV-irradiated seawater and model ligand (Table S1, Croot and Johannsson, 2000; Ardiningsih et al., 2020). Blanks were obtained by analysis in the absence of model ligands.

The application followed Abualhaija and van den Berg (2014) but used the above-described BASi instrument. SA (added to a final concentration of 5 µM), buffer, Fe additions (Table S1) and samples were left to equilibrate overnight.

For SA25, the buffer and DFe were added 1 h before analysis. SA was added to a final concentration of 25 µM separately to each vial, 15 min before the measurement.

All applications have been calibrated by reverse titration with DTPA. We would expect to recover comparable binding parameters for DTPA during the Fe titration. However, in all cases the logKcond for DTPA calculated from the Fe titration was overestimated. The overestimation of KFeDTPAcond for all three added ligands is likely a result of αFeDTPA<D and thus theoretically below the detection window for all applications. For determinations in marine samples, Caprara et al. (2016) showed in a compilation of data from the open ocean that, with the exception of NN, the ligands were above D of the used AL; thus deviation caused by ligands with α<D is likely a minor problem in seawater samples. For TAC and SA5, [L] was underestimated. Alt hough such a discrepancy in [L] could be a result of incorrect estimation of the Fe present in the titration, analysis with ICPMS showed that the Fe concentration increased only by 0.04 nM upon 2 nM DTPA addition, and thus we ruled out contamination as a cause for the underestimation of [L] for TAC and SA5. The Kcond values are comparable between the applications (ratios vary between 1 and 1.03, Table 3), although the range 21.3–21.8 is substantially higher than 19.0, the Kcond used for the calibration. Due to the codependence, the DTPA ligand concentration (2nM) should have been underestimated (Apte et al., 1988), which is the case for the results from TAC and SA5 (by a factor of 0.56–0.87, Table 3). But the DTPA ligand concentration was overestimated by SA25 (by a factor of 1.31, Table 3; see also titration in Fig. 1). Indeed, in the log-log plots (Fig. 2a) our results are well off from the theoretical titration line. Additionally, at the very low concentrations the data points of all applications deviate from the modeled curves.

The differences between the applications are also not large for phytic acid; the two SA applications are even quite similar. The large difference for the Kcond values of phytic acid estimated by TAC was not expected when comparing the two very similar titration curves. We found that small changes in determined FeTAC2 concentrations at low Fe additions could be responsible for this difference (Fig. 2b). When the calculation was repeated for the combined titrations, we obtained logKcond=22.2 (±0.2 and 0.2) and [L] = 1.6 ± 0.1 nM Eq.

The siderophores have high Kcond, so high that the AL should not be able to compete. However, although the here-estimated Kcond values are the highest compared to other model A ligands, curved titrations were still obtained (Fig. 1c, d, e), although there is considerable variance (Kcond=22.1–23.5, 21.4–22.9 and 20.2–21.7 for TAC, SA5 and SA25, respectively). The Kcond obtained for ferrichrome is close, almost identical, to those for desferrioxamine B for the three applications. However, [L] values obtained for ferrichrome are higher than found for desferrioxamine B, with a factor of 1.4–1.6. The Kcond values for desferrioxamine B are lower than the value calculated from thermodynamic stability constants (Hider and Kong, 2010; Schijf and Burns, 2016) (Tables 2 and 3). Although we want to focus on comparing the applications and not on the exact values, here we need to compare with literature values. The Kcond=20.2 for desferrioxamine B obtained for SA25 is much lower than measured by Rue and Bruland (1995), Kcond≥23, although they recovered 100 % of the added 2.5 nM. However, we note that they used another protocol and applied 4 min of nitrogen purging before every measurement, which would have interfered with the signal stability according to Abualhaija and van den Berg (2014). Buck et al. (2010) also successfully recovered 100 % of a different siderophore (aerobactin) using SA25. Witter et al. (2000) measured siderophores with CLE-AdCSV but using NN, and their results compare better with the results obtained here. They found a range of Kcond values for a range of siderophores and measured 21.6 for both desferrioxaine B and ferrichrome. These values are very close to those obtained here by SA5 (21.4–21.5). However, their Kcond for phytic acid was Kcond higher (22.3 with respect to Fe3+) than what we found with all the SA applications. However, other research reported lower Kcond for Fe–phytic acid complexes. Schlosser and Croot (2008) combined cross-flow ultrafiltration with the Fe radioisotope (55 Fe) and obtained a substantially lower value (18.6 with respect to Fe3+). Moreover, phytic acid can form colloids with FeIII (Anderson, 1963). Colloid formation will interfere in several ways with the analysis by the loss of Fe, since formation of colloids results in a potentially inert fraction of Fe, the loss of phytic acid and interference of the colloids on the mercury electrode surface. However, the formation of these colloids is dependent on the phytic acid concentration (Rijkenberg et al., 2006; Purawatt et al. 2007), and at 2 nM phytic acid we do not expect colloids to be formed. The ratios of added [L] and obtained values by CLE-AdCSV by Witter et al. (2000) varied between 0.8 and 1.7, resembling our results, although the ratio was 1 for both desferrioxamine B and ferrichrome. Thus, even for model ligands there is no consistence in the literature between ligands or methods, suggesting problems in the standardization of the methodology. It is possible that the siderophores used are not of 100 % purity, which would result in a systematic underestimation of [L]. However, whilst it is interesting to note absolute values, our research focuses on differences between the three applications, which should not be impacted by any impurities.

The theoretical titration curve for desferrioxamine B has a relatively large offset at low concentrations compared to the modeled results (Fig. 2c). This indicates that the theoretical Kcond was not even approached by the three applications. That we obtained (and not for the first time) clearly curved titrations, where we should not within the applied D values, is hard to explain. One possible explanation could be a reaction taking place at the electrode surface in CSV promoting ligand exchange of Fe(III) siderophore complexes, which produces a current. Another alternative explanation might be aluminum competition (which is not accounted for by the thermodynamic constants) since Al complexes with siderophores are detected in MS analysis of samples (Gledhill et al., 2019). The Al content, however, is unknown.

Ferrioxamine E was the only model ligand that was saturated with Fe prior to the start of the experiment. Moreover, none of the ALs should sequester Fe from ferrioxamine E, which is required in order to estimate Kcond because its Kcond is too high, outside D (Apte et al., 1988; Hudson et al., 2003; Gerringa et al., 2014). This can be explained by considering the Langmuir isotherm, used to derive Kcond and [L]: 5FeL=KcondL[Fe3+]KcondFe3++1, or 6FeLL=KcondFe3+KcondFe3++1, which shows that when Kcond×[Fe3+]=1, [FeL] = 0.5 [L], the equivalence point of the titration, where an almost linear relationship between [FeL] and [L] changes into an asymptotic relationship. In an Fe titration when Kcond[Fe3+] ≫ 1, [FeL] will approach [L]. When a titration starts at initial [FeL] > 0.5 [L], the ability to estimate Kcond diminishes substantially. In the asymptote, at much larger values of [FeL] / [L], the dependence of Kcond is lost and Kcond becomes impossible to derive. Therefore, the titration of ferrioxamine E was more or less a standard addition and in theory [L] should be equivalent to the determined DFe concentration. However, all three applications overestimated the ligand concentration, but the estimated [L] of ferrioxamine E compares very well for TAC and SA5 and even SA25. The relatively large difference in resulting Kcond between the two SA applications for the hydroxamate siderophores was not expected since the D values are close. The two most probable explanations are D and lack of equilibrium. Possibly the siderophores have αFeL at the borders of and greater than D, and therefore a small decrease in D still had a consequence for the outcome of the calculations. The short waiting time may be the other reason for the deviation of SA25, which we will discuss in the next section.

The [L] of the catechol vibriobactin was underestimated by SA5 and TAC and overestimated by SA25. We believe that this divergence was a combination of lack of equilibrium due to the high stability of Fe–vibriobactin complexes during the short equilibrium period of SA25 and consequent overestimation of [L] and possibly by the tendency of catecholates to oxidize or hydrolyze in water (Brickman and McIntosh, 1992), which could have resulted in partial loss of vibriobactin during overnight equilibration.

Our titrations of FA and HA show remarkable differences between the applications, [L] by TAC was 13 %–25 % of [L] by SA25 (Table 3). TAC detected 55 %–70 % of FA and HA in contrast to an early report that TAC could not detect any portion of IHSS humic reference material (Laglera et al., 2011). Our FA and HA results are in line with the partial detection of HS by TAC that was observed already by several field studies, where an increase in [L]TAC correlated with an increase in natural humics (Gerringa et al., 2017; Dulaquais et al., 2018; Slagter et al., 2017, 2019; Laglera et al., 2019). HS showed the largest deviations from the expected (literature) results of all tested ligands in [L]. Humic substances are ubiquitous in seawater (Laglera et al., 2009; Whitby et al., 2020; Yamashita et al., 2020) and potentially more representative of the dominant fraction of dissolved organic matter actually present in seawater than the model A ligands tested here, since although siderophores are detected in seawater, they are typically only present at pM concentrations (Mawji et al., 2008, 2011; Velasquez et al., 2016; Boiteau et al., 2018). The different results between the three applications could explain a major part of the offset between the TAC and SA methods in natural waters (Buck et al., 2012, 2015; Slagter et al., 2019; Ardiningsih et al., 2021). Moreover, the deviation in Kcond obtained by TAC from the other two applications (KTACcond/KSA5cond and KTACcond/KSA25cond up to 1.1) is greater than for most model A ligands (Table 3). This is also likely to be linked to the heterogeneity of humic substances, which means the detection window of each method will have a greater influence on the groups of binding sites titrated during the experiments. We cannot provide a definitive explanation for the [L] spread. TAC showed almost straight-line patterns for FA and HA (Fig. 1g, h), as in HS-rich estuarine waters (Gerringa et al., 2007; Croot and Johansson, 2000). This could be compatible with a fraction of HS being too strong and a fraction too weak to compete with TAC (both fractions would be at or beyond the upper and lower limits of D). There is an abundant presence of strong binding sites in HS that may not be outcompeted by TAC, since desferrioxamine B could also not outcompete all HS binding sites in Arctic Ocean samples (Laglera et al., 2019). Another possible explanation for the similar recoveries for FA and HA, despite their reported different affinity for iron, is that TAC could form interactions with some of the binding groups of HS, canceling their interaction with iron. In other words, the use of TAC would not obey Langmuir assumption 5. For the SA applications, [L] with SA25 seems to be substantially over the literature values in contrast to SA5. Titration data of HA with SA5 showed detectable levels of Fe–AL at low total Fe concentrations, while for SA25 they could not be seen. Thus, the formation of the electro-active Fe–SA complex does not happen until after 6–7 nM Fe has been added (0.1 mg HA, or over 10 nM for 0.2 mg HA). This is most probably an effect of ongoing association and dissociation processes between Fe, SA and HA, i.e., a lack of equilibration. Another explanation could be a substantial decrease in the sensitivity caused by adsorption of free and complexed humics onto the surface of the electrode, shielding the electrode from interaction with Fe(AL) complexes (Laglera et al., 2011, 2017). Adsorption of humics at the mercury electrode has been extensively discussed by Buffle and co-authors (Buffle and Cominoli, 1981; Cominoli et al., 1980), and the drop of sensitivity for CLE-AdCSV was discussed in Laglera et al. (2011 and 2017). The SA25 application does not obey Langmuir assumption 1. This might also explain the large data spread in Fig. 2h for SA25.

The log-log plots for DTPA and desferrioxamine B, between known and observed conditional stability constants, show that the data points obtained by TAC are closest to the theoretical curve of DTPA, and those obtained by SA5 are closest to desferrioxamine B (Fig. 2a, c). However, the TAC application has the highest D and should therefore be better equipped to detect desferrioxamine B and least equipped to detect DTPA. The different results between applications are mostly due to data in the first curved part of the titration as shown in Fig. 1 and illustrated when compared with the theoretical titration curves. At this part of the titration curve, peaks should in many cases be below the detection limit. Thus the precision of these measurements is very low. The log-log plots (Fig. 2) emphasize the differences between expected and observed values. The observations seem to overestimate the FeAL at low metal additions. Possible reasons are

electrochemical, for example a catalytic effect becoming more important at low concentrations and enhancing the signal or tiny peaks caused by impurities of the reagent or the methanol solvent as shown in previous work with NN (Boye et al., 2001),

It is also possible that reactions occurred during the cathodic scan, which could also explain the deviating results of [L] for vibriobactin, which was underestimated by 52 %–62 % using TAC and 60 %–70 % using SA5. Free catechols can be electro-active (Fakhari et al., 2008), and even a small contribution to the CSV peak from the ligand side of the complex would lead to a significant underestimation of the complexed fraction. A last explanation of underestimating ligand concentrations can be contamination after sampling for Fe determination by ICPMS took place.

Considering average [L] and the spread in [L] in the titrations with 2 nM of model A ligands (without considering the saturated ferrioxamine), average [L] was 1.51 ± 0.32, 3.30 ± 0.76 and 1.96 ± 0.73 nM Eq Fe for TAC, SA25 and SA5, respectively. There appeared to be model-ligand- and AL-dependent variations in the estimation of [L] as also illustrated in Fig. S6. We can conclude that TAC underestimated most added model ligand concentrations, with a model-ligand-dependent degree of underestimation between 0.55–0.7 for HS and an average of 0.8 (0.52–1.05) for the model A ligands. The application with SA5 both over- and underestimated model ligand concentrations but less than SA25 and TAC, respectively, although HS was overestimated by a factor of 1.27–1.82 (assuming the number of binding sites from literature). The application with SA25 overestimated the concentrations by a factor of 1.3–2.3 for model A ligands and 2.33–4.17 for HS. However, it must be remembered that [L] and Kcond are not determined independently, and unfortunately, comparison of thermodynamic constants with our results suggests that Kcond cannot be estimated precisely. Here SA applications result in worse estimates of Kcond due to a lack of data at FeL < L and the possibility that the ligands are outside D, the detection window. As far as we know, four publications describing an intercomparison exist. Three of these compared SA25 and TAC (Buck et al., 2012, 2016; Slagter et al., 2019) and one compared SA5 and TAC (Ardiningsih et al., 2021). In all publications, [LSA] is larger than [LTAC] when the data are fitted with a one-ligand model. In Buck et al. (2016) it was concluded that, when using the same calculation method, comparison between the results of both applications seemed good, with one exception that SA could measure a second ligand whereas TAC could not, and therefore the total ligand concentration obtained with SA25 was always considerably larger (their Fig. 2e). This difference was attributed to an underestimation by TAC because TAC does not detect binding sites of humic substances and cannot discriminate a second ligand as well as SA25 (Buck et al., 2012, 2016; Slagter et al., 2019). Slagter et al. (2019) sampled in the Arctic Ocean where the Transpolar Drift transports high concentrations of humic substances in the upper 50 to 80 m. Since the humic content was an important feature, TAC was compared with SA25 (Slagter et al., 2019) and the voltammetric determination of humic acids (Sukekava et al., 2018). Slagter et al. (2019) found [LTAC/[LSA25]=0.6. However, this ratio hardly varied with the concentration of humic substances, a strong indication that the underestimation of humics with the TAC method was not the only explanation for the difference in ligand concentration between the two methods. Ardiningsih et al. (2021) compared TAC and SA5 in the Arctic Fram Strait and also concluded that the offset between TAC and SA5 could not only be directly ascribed to underestimation of binding sites in humic substances. Moreover, Ardiningsih et al. (2021) found a relatively constant [LTAC/[LSA5] on the Greenland shelf but a variation in [LTAC/[LSA5] between 0.6 and 1 in Fram Strait. It was largely the inconsistencies in these studies, where humic substances were believed to potentially play a key role in Fe speciation, that led to this study. In future use of CLE-AdCSV, careful consideration is needed for the interpretation of the obtained [L] in relation to the application and environmental variations in ligand groups, especially the humic substances.

No intercomparison between SA5 and SA25 has been undertaken since the SA5 application was published in 2014. The question remains as to why SA5 and SA25 in this work give different results. One explanation might be disequilibrium of the SA25 application. To further study the equilibration process, we executed some kinetic experiments.

In-cell kinetics were first performed with three types of samples: normal seawater, UV-irradiated seawater and UV-irradiated seawater containing DTPA at large concentrations (200 for TAC and 40 for both SA concentrations) (Fig. 3). Further, in-cell kinetic experiments were done on a subset of the model ligands at lower concentrations (2 nM or 0.2 mg). For the model A ligands, DTPA was chosen since it was used as the calibrating ligand. Desferrioxamine B was chosen to represent the hydroxamates, and vibriobactin was chosen to represent the catecholates. Phytic acid was also included because the titration results of all applications were in agreement. FA was used to represent model B ligands, as heterogenous natural organic matter.

The in-cell kinetic measurements with high DTPA gave completely different results between TAC and SA (Figs. 3 and 4, Metrohm stand used). For SA25, the peak decrease was high at t=3 min (initial value) and dropped sharply to values close to the limit of detection within an hour in UV-irradiated seawater and UV+DTPA. For TAC we observed a slow increase in the Fe(TAC)2 concentration followed by an asymptotic change to a constant value, as expected for a product of a ligand exchange reaction tending towards equilibrium.

Kinetic experiments with the low concentrations of model ligands in a volume of 10 mL were not pursued with the SA applications, only with TAC. In these experiments, equilibrium between TAC and model ligands was reached after approximately 6 h, as observed previously (Croot and Johansson, 2000), 4 h for most ligands and 6 h for desferrioxamine B (Fig. S7). Although the rate of Fe(TAC)2 formation changed with the type of model ligand, a steep increase where the relative weaker ligands were added, DTPA, FA and phytic acid, compared with a slow and steady increase where desferrioxamine B and vibriobactin were added, all model ligands followed the theoretical ligand exchange concentration evolution (Fig. S7).

Possible explanations of the rapid decrease in peak height are as follows.

The decrease can be due to formation of the non-electro-labile Fe(SA)2 complex. Fe(SA)2 is the non-electro-active species (Abualhaija and van den Berg, 2014) and becomes the dominant Fe(SA)x complex at higher SA concentrations. The two forms of Fe–SA would have different formation kinetics, with a slower formation of Fe(SA)2, with Fe coming not from the dissociation of the model complex but from the dissociation of Fe(SA). This process increases the time to reach equilibrium, and D changes accordingly. We monitored the electro-labile Fe(SA) concentrations after SA additions in the range 2.5 and 50 µM using the adapted Metrohm instrument with a small mercury drop (size 1) and regular air purging (Fig. 4). Possible contributions due to decreasing oxygen were excluded and due to adsorption on mercury on the cell bottom were minimized. At 25 µM SA the concentration of the electro-active species practically disappeared after 2 h. At SA < 25 µM, the concentration of the electro-labile species Fe(SA) decreased exponentially with time for a period of at least 13 h. At concentrations ≤5 µM there is a decrease to a constant value above zero. These results support the formation of a non-electro-active species Fe(SA)2 irrespective of adsorption on the mercury drop, confirming the results of Abualhaija and van den Berg (2014).

The kinetic experiments were repeated extracting 10 mL aliquots from a 200 mL bottle. Experiments carried out were UV-irradiated seawater, UV-irradiated seawater with desferrioxamine B, UV-irradiated seawater with phytic acid and UV-irradiated seawater with FA (Fig. 6). Some points have to be considered for interpretation. The conditioning procedure of the bottle in which the reaction takes place raises the question of whether to condition with or without the AL. Addition of AL should occur at t=0, and conditioning of the bottle with AL is thus not possible. Conditioning without AL with UV-irradiated seawater with 6 nM of added DFe could cause Fe precipitation and/or adsorption on the bottle walls. DFe at the end of the experiment is probably higher than 6 nM due to Fe desorption from the bottle wall.

The results with TAC showed the same pattern as for the in-cell experiments: an increase that levels off to equilibrium for UV-irradiated seawater, phytic acid and FA (Fig. 6) characteristic of a ligand exchange reaction reaching a steady state. Higher equilibrium signals prove an extra Fe input from the bottle wall. The experiment with desferrioxamine B did not level off after 8 h, and the slope was less steep. This concurs with our in-cell observation that desferrioxamine B dissociates more slowly. That equilibrium is not reached even after 8 h explains, at least for this model ligand, the underestimation of [L] by TAC (Table 2; Croot and Johansson, 2000). It also indicates that with this protocol kinetics were dependent not just on ligand exchange but also on Fe desorption and/or redissolution of precipitated Fe.

The results of the bottle experiments for SA5 changed dramatically with respect to those of the in-cell protocol. They show that equilibrium appears to be achieved for phytic acid and fulvic acid, and the increase in signal over time in the first 1–2 h is similar to that for TAC (Fig. 6). An equilibrium is reached after approximately 4 h with phytic acid and FA. More time, 8 h, is needed to reach an equilibrium with desferrioxamine B, in line with the slower formation of FeTAC2 in the presence of desferrioxamine B compared to the other model ligands (Figs. 6, S6).

Equilibrium or a steady state is difficult to establish for SA25. A plateau is reached after approximately 1 h. This seemingly depends on the added ligand, but after the plateau the signal decreases at a steady rate. We can conclude thus far that the steep decrease shown in the BASi electrode in the in-cell kinetics does not happen with the bottle experiments. This result points to adsorption on the dispensed mercury puddle as the main cause of the disappearing signal. Equilibrium is not reached at short waiting times after addition of 25 µM SA. Therefore, the use of the Langmuir isotherm to calculate the ligand characteristics is not possible. We hypothesize that the distinction in more than one ligand group, which was often possible with this application, could have been caused by the absence of equilibrium.