Sample solutions were composed to mimic the halide composition of seawater. This was achieved using either dissolved Instant Ocean (Spectrum Brands) or commercially available halide salts at a composition that mimics Instant Ocean (for consistency) in solutions referred to hereafter as “saltwater”. The halide concentrations in these solutions were made to a final concentration of 0.56 M Cl-, 7.2 × 10-4 M Br-, and 1.9 × 10-6 M I-. Except for Instant Ocean, all chemicals were purchased from Sigma Aldrich. Halide salts include solid NaCl (puriss. p.a. grade, ≥99.5 % purity), NaBr (puriss. grade, > 99 % purity), and KI (puriss. p.a. grade, ≥ 99.5 % purity). We note that these halide concentrations are comparable to those in actual seawater (Herring and Liss, 1974; Luther et al., 1988; Tsunogai and Sase, 1969), which typically contains Cl-, Br-, and I- at ratios of 1:1/660:1/200000. Solutes were dissolved in ultrapure water (Birck Nanotechnology Center). Dissolved organic carbon for Instant Ocean and halide salt solutions was analyzed using a Shimadzu TOC-VCSH Total Organic Carbon Analyzer and determined at approximately 70 mg L-1 for Instant Ocean solutions and less than 5 mg L-1 for saltwater solutions. No further characterization of carbon-containing compounds was performed.

While previous investigators have adjusted the pre-freezing pH of their samples, it is very difficult to know the pH in the surface brine (or disordered interface) of frozen samples (Bartels-Rausch et al., 2014), though there is evidence from laboratory studies suggesting that the pH of salt solutions remains largely unchanged after freezing (Wren and Donaldson, 2012b). To obviate this problem, the aqueous solutions used in this study were buffered so that the same pH should exist in the surface brine layer. All solutions were buffered by either a 20 mM acetic acid (ACS reagent grade, ≥99.7 % purity)/acetate (puriss. p.a. grade) buffer (pH ≈ 4.5–4.7) or a 20 mM bisulfate (ReagentPlus grade, 99 % purity)/sulfate (ReagentPlus grade, ≥99.0 % purity) buffer (pH ≈ 1.7–2.2). These buffer concentrations were chosen as a compromise between using as little buffer as possible, yet enough buffer to ensure adequate buffering ability, as buffer capacity rapidly decreases as constituent species concentrations approach the acid Ka value. pH values of sample solutions were determined before and after experiments with no significant changes observed, suggesting the buffer composition/buffering capacity does not appreciably change over the course of an experiment (discussed further in the Supplement). 100 µM of either hydrogen peroxide (trace analysis grade, ≥30 % purity) or sodium nitrite (ReagentPlus grade, ≥99.0 % purity) was included as photochemical hydroxyl radical precursors, via Reactions (R5)–(R7).

Experiments were performed in a custom-built 150 cm long, 2.5 cm ID frozen-walled Pyrex flow tube contained within a temperature-controlled cooling jacket. In each experiment, 80.0 mL of sample solution was poured into the tube in the presence of room air, which was subsequently sealed with vinyl caps (McMaster-Carr). The flow tube was then rotated on motorized rollers within a 170 cm × 50 cm × 50 cm insulated wooden cooling chamber. Crushed dry ice was placed along the bottom of the chamber, and fans were used to circulate the air throughout the chamber such that the flow tube was evenly cooled. After ∼30 min, the sample was evenly frozen (ice thickness of 0.9 mm). The flow tube was subsequently transferred to an enclosed 156 cm × 50 cm × 50 cm wooden Mylar-lined experiment chamber and connected to a recycling chiller set to 258 K (i.e., above the NaCl⚫2H2O eutectic point. At this temperature, the relevant chemical reactions are expected to occur with/in a brine on the ice surface; Cho et al., 2002; Oldridge and Abbatt, 2011). This conjecture is based on the work of Oldridge and Abbatt (2011), who reported from a series of similar experiments that when O3 is flowed over frozen NaCl/NaBr solutions above the NaCl eutectic temperature, reaction kinetics were strongly consistent with chemistry occurring in a liquid brine. The cooling liquid used for the chiller was a mixture of 60 % ethylene glycol and 40 % distilled water. Six UVA-340 solar simulator lamps (Q-Lab, 295–400 nm with maximum wattage at 340 nm, irradiance spectrum in Fig. S1 in the Supplement) were installed in the experiment box (two on each side except bottom). Each side was lined with reflective Mylar sheets to evenly irradiate the flow tube when the lamps were powered.

A flow schematic representing typical experiments is shown in Fig. 1. The carrier gas (Air, Ultra Zero grade, Praxair) was scrubbed of volatile organic compounds using activated charcoal and water by traveling through coiled stainless-steel tubing surrounded by crushed dry ice (replaced throughout the course of an experiment). This gas was measured to contain ≤300–400 pmol mol-1 NO (experimentally determined limits of detection) using the Total REactive Nitrogen Instrument (TRENI) (Lockwood et al., 2010; Xiong et al., 2015). Though NO2 was not measured, it should have been removed by the charcoal trap. Before entering the coated-wall flow tube, the carrier gas flowed through a commercial O3 generator (2B Technologies model 306). Carrier gas air entered the tube near room temperature (20 ∘C). At the start of experiments, the O3 generator was set to 0 nmol mol-1. Carrier gas then entered the flow tube in the dark experiment chamber. In most experiments, the carrier gas was regulated to a volumetric flow rate of 4.0 L min-1, which yields a residence time in the flow tube of ∼12 s. On exiting the flow tube, sample air was characterized using a Thermo Environmental 49i O3 monitor (flow rate of ∼1.5 L min-1) and a chemical ionization mass spectrometer (CIMS; sampling flow rate of ∼1.7 L min-1; described below in Sect. 2.3). Excess flow air was vented away. At set times in an experiment, the solar simulator bulbs were activated, and O3 was added to the system by powering the O3 generator. At the end of each experiment, the ice was melted and the water collected for pH measurements. To clean the flow tube, its interior was washed three times with ultrapure water before a final rinse with wash acetone. The flow tube was then connected to a compressed nitrogen gas cylinder (Praxiar, > 99.99 % purity) to dry for at least 2 h. Once dry, the flow tube was disconnected and capped until the next experiment.

Halogen species were detected using a chemical ionization mass spectrometer (CIMS), described previously by Liao et al. (2011) and Pratt et al. (2013). Chemical ionization is achieved by ion-molecule reactions that occur between iodide–water reagent clusters, I(H2O)n- in N2, and the gas-phase analytes in zero air. The iodide–water clusters are formed when gas-phase iodide ions, generated by flowing 5 ppm methyl iodide through a 210Po ionizer (NRD), combine with water in the humidified ion-molecule region of the CIMS. Ion were filtered using a quadrupole mass filter. The ice-coated flow tube was connected to the CIMS via approximately 50 cm of i.d. 1/2 in. PFA Teflon tubing.

A typical CIMS sampling cycle consisted of an 8.35 s duty cycle. Dwell times for all monitored species were 250 ms except for the reagent ion (detected as m/z 147, I(H218O)-), which was set to a dwell time of 100 ms. The 18 ions analyzed in this study are listed in Table 1, but we focus herein on results concerning masses related to Br2 (m/z 285 and 287: I79Br79Br- and I81Br79Br-, respectively), Cl2 (m/z 197, 199, and 201: I35Cl35Cl-, I37Cl35Cl-, and I37Cl37Cl-), and I2 (m/z 381: I3-). In addition, IBr (m/z 333 and 335: I79IBr-, I81IBr-) was unambiguously detected in some experiments. The presence of Br2, Cl2, and IBr was confirmed by measuring the ratios between the two isotope signals for each mass, compared to the natural abundances (i.e., 1.95 for m/z 287:285; 1.54 for m/z 197:199; and 1.03 for m/z 333:335, respectively). Data outside ±25 % the expected isotope ratio were excluded from analysis. The signals for BrCl (m/z 241 and 243: I79Br35Cl-, I81Br35Cl-, I79Br37Cl-) masses were never observed at the correct ratios (1.3 for m/z 243:241), and so those data were not reported here. As the introduction of ∼60 nmol mol-1 O3 to the experimental system significantly increased the baseline signal of m/z 197, but not m/z 199 or 201, the presence of Cl2 could not be confirmed under elevated O3 conditions. In addition, background-subtracted, relative signals for m/z 271 (IHOI-) and m/z 225 (IHO81Br-) are discussed (signals are relative to that of the ionization gas (m/z 147, I(H218O)-). According to isotope ratios, IHOBr- was not unambiguously observed, however, due to an interference at m/z 223 (IHO79Br-), and our results here should be considered for only qualitative purposes as we only discuss relative changes in the signal.

CIMS calibrations were performed using I2, Br2, and Cl2 permeation devices (VICI) at the start and conclusion of each experiment. Br2 and Cl2 permeation outputs were quantified using the spectrophotometric method described by Liao et al. (2012). The I2 permeation output was quantified by flowing the I2 through an impinger containing a NaHCO3 (30 mM)/NaHSO3 (5 mM) reducing solution. This solution quantitatively reduces I2 to I-, which was then determined using a Dionex DX500 ion chromatography system. Permeation rates were calculated for each experiment and found to average (1.9±0.1) × 10-11, (5.5±0.1) × 10-10, and (8.6±0.1) × 10-10 mol min-1 of I2, Br2, and Cl2, respectively (uncertainties representing standard error of the mean). CIMS calibration factors were calculated for individual experiments. These factors are based on the average of the signal sensitivities, determined from the permeation sources, calculated at the start and completion of each experiment. Corresponding uncertainties for these calibration factors thus represent the 1σ standard deviation of the mean sensitivity. An approximate I79IBr- calibration factor was assumed to be the average of the sensitivities for m/z 287 (IBr2-) and 381 (I3-). Background measurements were performed before and after the experiment (minimum of 5 min) by passing the carrier gas through the experimental flow tube (without O3, in the dark) and subsequently through a glass wool scrubber, previously shown to remove molecular halogens with greater than 95 % efficiency (Liao et al., 2012; Neuman et al., 2010). Temporal variations in bromine-species signals while using the low pH sulfate/bisulfate buffer were observed in some experiments (Fig. S2) and are discussed in the Supplement.

Analysis of experimental data was based on 1 min averages, with uncertainties representing the standard deviation of these averages. Subsequently, signals were converted to concentrations using the sensitivities calculated above, propagating the sensitivity uncertainty into the measurement uncertainty. Average limits of detection (3σ) across all experiments for the molecular halogens during background periods were 1.8±0.4, 1.2±0.3, and 9±2 pmol mol-1 for Br2, Cl2, and I2 respectively (uncertainties representing standard error of the mean). Additionally, reported uncertainties for integrated amounts of formed halogens are calculated as integrated halogen concentrations multiplied by the relative uncertainty in the CIMS signal sensitivity.

After the initial connection of the flow tube to the CIMS (i.e., before irradiation or addition of O3), large I2 signals (measured as I3-, m/z 381) were observed in several cases in which OH-radical precursors were utilized, especially when pH ≤ 2 (e.g., Figs. 2b and S2). Integrated calibrated sums of this dark I2 production are estimated in Table S1 in the Supplement and span from the time when the flow tube was connected to the CIMS until lights were activated. When pH ≤ 2, dark production of I2 sometimes caused significant depletion of reservoir I-. Experiment IO4 and SW5 (both using H2O2 as an OH precursor) only had, at most, ∼36 % of the initial 152 nmol of I- by the time lights were turned on (remaining I- was estimated by subtracting twice the observed I2 (i.e., two I- for every I2) from the initial 152 nmol of I- in the IO or SW solutions). Considerably less dark I2 production occurred using NO2- as an OH precursor (depleting I- by an average of 4.5 %; Table S1). However, the amounts in Table S1 represent lower limits of the dark-produced I2; it is impossible to accurately determine the extent of dark I2 production since some was lost from the flow tube during its connection to the CIMS after freezing (Fig. 1). At pH ≈ 4.7, this production was relatively modest. Only Experiment IO2 was noticeably affected, in which only ∼0.5 % of initial I- was removed by dark mechanisms (Table S1). Under both pH regimes (i.e., ∼4.7 and < 2), this signal subsequently decayed as I2 flushed out of the system until a low steady-state concentration was reached. No corresponding dark production of Br2 or Cl2 was observed for any experiments at any pH.

As previously reported, both H2O2 and NO2- can directly convert I- to I2 under dark acidic conditions. The oxidation of I- by H2O2 occurs through the condensed-phase reactions, Reactions (R23) and (R24) (Küpper et al., 1998): R23I-+H2O2↔HOI+OH-R24HOI+I-+H+→I2+H2O. Nitrite ions react with hydronium ions to form the nitroacidium ion, H2ONO+, which has been previously shown to produce I2 (Reactions R25–R27) (Hellebust et al., 2007; O'Driscoll et al., 2006, 2008; O'Sullivan and Sodeau, 2010): R25NO2-+H3O+↔HONO+H2OR26HONO+H3O+↔H2ONO++H2OR272H2ONO++2I-↔2NO+I2+2H2O.

Therefore, it is likely the I2 observed on connection of the flow tube to CIMS originated from the above reactions, Reactions (R23)–(R27), as the pH ≤ 2 experiments in this work (IO3-5, SW3-5) favor these forward reactions.

In experiments without an OH source (IO6–IO8, SW6–SW8), I2 production was greatest when O3 was introduced to the irradiated tube for both pH regimes (Table 2). The amount of I2 produced over 60 min in these experiments was large, ranging from 26±9 to 80±1 nmol at pH = 4.7 and from 2.6±1.7 to 38±12 nmol at pH < 2. This production likely results from a combination of heterogeneous recycling and the surface and aqueous reactions between O3 and I- (k=2.0×10-12 cm3 molecules-1 s-1; Liu et al., 2001). While the I2 produced when pH < 2 appears to be lower, I2 had already been produced in the presence of light prior to addition of O3 (Sect. 3.2.2), yielding a lower [I-]/[Br-] ratio when O3 was eventually added. Br2 production amounts ranged from 0.012±0.001 to 0.16 ±0.01 nmol at pH = 4.7 and took up to 6 h to rise above detection limits after O3 was added. At pH ≤ 2, Br2 production amounts ranged from 0.14±0.02 to 0.93±0.05 nmol. While O3-mediated halogen production has been observed directly from frozen surfaces in the absence of light in previous laboratory studies (Artiglia et al., 2017; Oldridge and Abbatt, 2011; Oum et al., 1998a; Wren et al., 2013), Br2 production has not been directly observed from the Arctic snowpack without irradiation (Pratt et al., 2013). This raises a question of the role of O3 in initial halogen release in the Arctic spring.

When OH precursors were present, the addition of O3 to the zero-air flow over the irradiated frozen sample caused additional production of I2 and Br2, as shown in Fig. 2a and b, under both pH regimes (Table 2). In experiments at pH ≈ 4.7, in which [I-]/[Br-] remained sufficiently large due to minimal dark production of I2 (i.e., IO1-2, SW1-2), exposure to O3 caused a sharp increase in I2 (as in Fig. 2a). I2 production amounts for frozen Instant Ocean at pH ≈ 4.7 (IO1, IO2) averaged 22±10 nmol, about 2 times less than for frozen saltwater experiments SW1 and SW2 (average production amount of 51±25 nmol). As the I2 signal decayed, the corresponding Br2 signals gradually increased above detection limits, approximately 3 h after the introduction of O3 (Fig. 2a). The average integrated amounts of Br2 produced from these pH ≈ 4.7 experiments were very similar (0.05±0.01 nmol for IO experiments and 0.03±0.01 nmol for SW experiments).

When pH < 2, the effects of O3 addition varied according to the remaining availability of I-. When the surface I- reservoir had been reduced from dark reactions with H2O2 or NO2- (Reactions R17–R21; Sect. 3.1), exposure to O3 did not increase I2 above the LOD except in Experiment IO5, which exhibited a small spike before decaying below the LOD (0.11±0.06 nmol in IO5). However, O3 did cause additional Br2 production after 1 h (average of 10±2 nmol for IO4 (Fig. 2b) and IO5 (Fig. S4) and 14±2 nmol for SW4 and SW5). In contrast, for SW3 (using NO2- as an OH source), there was relatively little initial consumption of I- by dark reaction; therefore, when O3 was added, 1.1±0.1 nmol additional I2 was observed, comparable to what was observed with the higher pH experiments (Fig. S5). The amount of Br2 produced (0.46±0.01 nmol) was also significantly less than observed when I- was initially depleted, demonstrating the importance of the halide ratios.

This additional O3-induced halogen production could result from a combination of mechanisms. First, as discussed above, O3 can react with halides on frozen saline surfaces to produce Br2 or I2 per Reactions (R18)–(R19) and then Reaction (R4) (Artiglia et al., 2017; Carpenter et al., 2013; Gladich et al., 2015; Hayase et al., 2010; Oum et al., 1998a; Shaw and Carpenter, 2013; Wren et al., 2013). It is possible that Br2 (as well as other halogens) may have been produced via this mechanism at levels below the LOD in previous Arctic snowpack studies (Custard et al., 2017; Pratt et al., 2013; Raso et al., 2017).

The presence of O3 also yielded HOX compounds (Figs. 3–4), likely formed in the flow tube in part by O3 reactions with halides (Reactions R18–R19). Additionally, given a flow tube residence time of 12 s, gas-phase production of HOX is possible via Reactions (R1)–(R3) and could act as an additional X2 production source (via Reaction R4), given a timescale for molecular diffusion of 6.5 s for HOBr from the center of the tube to the ice surface. At this flow rate, there is enough time for one–two heterogeneous reaction cycles. Figure 3 shows HOX for IO2 (pH = 4.7 with H2O2 present, analogous to IO1, SW1, SW2). For each experiment in this series, increases in I2, HOI, and Br2 were readily observed when the O3 was introduced at hour 2 (Figs. 3 and S4). However, corresponding HOBr production was not observed, perhaps either due to a high LOD or the relatively low abundance of Br2 that would limit production of HOBr. Conversely, in pH ≤ 2 cases when substantial portions of I- had already reacted prior to irradiation (IO4, IO5, SW4, SW5), the addition of O3 produced negligible amounts of I2 and HOI (Fig. 4). But, in these cases, following the addition of O3, HOBr (m/z 225 IHO81Br-), was observed together with Br2 (Figs. 4 and S4). We note in this case that m/z 223, representative of IHO79Br-, does not appear to show an enhancement when O3 is added to the system. There was a much higher background signal for m/z 223 compared with m/z 225 (IHO81Br-) resulting from an unknown interference.