Surface seawater from 1.8 ± 0.3 m depth was routinely sampled from the station L4 of the Western Channel Observatory, located ∼ 13 km southwest of Plymouth, U.K (50.3° N, 04.22° W, full water depth ∼ 55 m) by the RV Plymouth Quest using 10 L Niskin bottles mounted on a rosette sampler, alongside a Seabird 19 + CTD. Sub-samples were transferred directly into overflowing acid-washed 5 L glass bottles, sealed with ground glass stoppers with no headspace and transported to the laboratory in a dark cool box. Experiments were carried out between 3 and 24 h after sample collection to minimize biological or chemical transformation of DOM. No significant relationships were observed for the duration between seawater sampling and VOC production measurements. Measurement campaigns were undertaken over contrasting seasons: late winter/spring (14 February–3 May 2017), summer/autumn (15 June–12 October 2020) and autumn/winter (4 September–5 November 2018). Chlorophyll a concentrations measured by fluorometry (Fig. 3c), and cell counts for total eukaryotes (Fig. 3d) and heterotrophic nanoflagellates and bacteria (Fig. 3e) measured by flow cytometry were sampled, processed and analysed according to the Standard Operating Procedures of the Western Channel Observatory (https://www.westernchannelobservatory.org.uk/data.php, last access: 4 November 2026).

Unlike previous ozonolysis studies that rely on quiescent or gently mixed waters (Schneider et al., 2019; Kilgour et al., 2024; Schneider et al., 2024), our approach uses a bubble-column system designed to emulate the high-surface-area, highly dynamic conditions produced by wave breaking in the surface ocean. This provides a complementary perspective because (1) bubbles generated during wave breaking entrain atmospheric O₃ and can deliver it directly into the near-surface water column, and (2) bubbles are known to efficiently scavenge surface-active organic matter from the water column and concentrate it at the gas–water interface. Our setup therefore enables investigation of O₃–seawater reactions under conditions that more closely mimic bubble-rich marine environments.

Ozonolysis experiments were carried out using an acid-washed, custom-built 1.45 L cylindrical borosilicate glass column fitted with a fine glass frit, beneath which a gas inlet entered a ∼ 0.2 L void. Silicone tubing was used to siphon 1.2 L of water sample into the column above the frit, minimising bubble formation and contact between the sample and lab air. A headspace of 0.25 L was maintained in the column above the sample, which also contained the gas outlet.

We briefly describe the experimental methods here, with detailed methods given in the Supplement. A mass flow controller (MFC; Bronkhorst ELFLOW Select series) was used to deliver a constant flow (150 sccm) of zero air (BOC BTCA 178 synthetic air) further scrubbed of organics using a custom-built Pt catalyst heated to 450 °C at 150 standard cubic centimetres per minute (sccm) to the gas inlet beneath the frit, resulting in a dense and turbulent plume of bubbles passing through the sample. The bubble interface area (Abub) was estimated to be 360 cm² from Abub=6Vbub/dbub, where Vbub represents the total bubble displacement volume (∼ 3 mL) and d approximates the diameter of a single bubble (∼ 0.5 mm). Vbub was measured by observing the volume displacement with and without the gas flow, while dbub was estimated photographically using a ruler as a reference. The bubble interface area is on the order of six times the cross-sectional area of the glass column, and the bubble residence time in the water phase was ∼ 1.2 s.

Ozone was generated from the artificial air using a corona discharge (Enaly 1000BT-12 ozone generator) and mixed into the zero-air stream feeding the experimental column. Mean ozone input mixing ratios differ between experiments conducted in spring, summer, and autumn – 7.7 ± 0.7, 1.7 ± 0.7, and 2.1 ± 1.1 ppmv, respectively. These inputs are substantially higher than typical ambient marine ozone levels (by approximately 25–150 ×) but were intentionally used to obtain (1) strong VOC production, and (2) high signal: noise ratio for the ozone measurement under the experimental constraints. Gas-phase ozone in the column headspace was monitored every 10 s using a dual-channel ozone analyser (2B Technologies Model 205). Because the analyser requires a relatively high inlet flow (∼ 2 L min⁻¹), a large make-up flow of zero air (dilution factor of ∼ 12) was added to meet the instrument's flow requirements, hence the need for high ozone input. The VOC production vs. ozone input relationship was experimentally determined to be linear at ozone input mixing ratio below ∼ 4 ppmv (see Fig. S2 in the Supplement). Gas-phase VOC concentrations in the outflowing air from the column headspace were quantified using a proton-transfer-reaction mass spectrometer (PTR-quadrupole-MS, Ionicon Analytik), as described in Phillips et al. (2021). The PTR-MS was operated in multiple ion detection mode, mainly monitoring m/z 45 (acetaldehyde), m/z 59 (acetone/propanal), m/z 63 (DMS) and m/z 69 (C₅H9+). The quadrupole mass dwell time was set to 50 ms for H₃O⁺ and 500 ms for each VOC. Overall, the temporal resolution of the measurement was 3.1–7.7 s (0.13–0.27 Hz), depending on the number of ions monitored. A certified VOC gas standard (Apel-Riemer Environmental Inc.) was used to routinely calibrate the PTR-MS. The limit of detection (LOD) was defined as 3σ of repeated blank measurements following Wohl et al. (2019) and LOD values for m/z 45, m/z 59, and m/z 69 are shown in the Supplement. Full details of PTR-MS data processing are shown in the Supplement.

Each experiment comprised four stages, some of which were repeated as the experiment progressed (Fig. 1). In stage one (S1), the ozone generator was on, and the bubble column was bypassed to measure the ozone input. In stage two (S2), the ozone generator was off, and the bubble column was bypassed to measure the dry VOC blanks. In stage three (S3), with the ozone generator remaining off, zero air was passed through the water sample in the column, resulting in the transfer of dissolved VOCs from water phase into gas phase and flushing the column headspace. After 20–30 min, measured VOCs stabilised, and the sample air was assumed to be in approximate steady-state with the seawater sample. In stage four (S4), ozone was added to the zero-air stream, and after ∼ 1–2 min measured VOCs started to increase. In contrast, ozone only became measurable ∼ 5 min after ozone addition. VOC mixing ratios peaked sooner than ozone because ozone was lost entirely at the beginning to DOM and other aqueous compounds (e.g. I⁻). We compute the VOC peak height by taking the difference between the steady-state point during the latter phase of stage three and the peak maximum during S4. We compute the VOC peak gradient by linear regression through the initial linear portion of the peak during stage four. Typically, two Milli-Q samples were run to check for column cleanliness and also provide a humid VOC blank, followed by triplicate seawater samples. Ozone and VOC were measured continuously throughout all four stages.

To explore potential biological VOC precursors and determine oxidation products, specific reactants were diluted in aged VOC-free seawater (aged in dark, gas-tight glass bottles for ∼ 1 month at 20 °C) and added to the bubble column, using the same experimental procedure as described for natural seawater above. Additions included: (1) Axenic algal culture (E. huxleyi Rutgers 607) which had been grown to senescence, added directly to the column by briefly removing the column cap and pipetting into the water sample (5–20 mL); (2) NOM sourced from the International Humic Substances Society (Suwannee River), dissolved in 10 mL of aged seawater and poured into the column (0–10 mg L⁻¹ column concentration); For each of these reactant addition experiments, the concentrations of the reactants were diluted to be of the same order of magnitude compared to the surface ocean dissolved organic carbon (DOC) concentration (ca. 100 µmol C L⁻¹).

VOC production from the ozonolysis of natural surface seawater showed clear seasonal variability that corresponded closely to the seasonal phytoplankton dynamics of the Western English Channel. Figure 2 shows the VOC peak height (Fig. 2A) and the peak gradient (Fig. 2B) following ozonolysis for the three most abundant ions detected by the PTR-MS. In late winter/early spring (DOY 45–80) the VOC response was stable at around 3 ppbv and 0.02 ppbv s⁻¹ for all compounds. DOY 80 onwards saw a more pronounced response to ozonolysis, with a rapid rise in the peak height of acetone/propanal (m/z 59), alongside a more gradual rise in peak height of m/z 69 and acetaldehyde (m/z 45) (Fig. 2A). This period saw rises of 65 %, 136 % and 348 % in peak heights of m/z 45, 59 and 69, respectively, compared to the weeks preceding DOY 80. Peak gradients of m/z 69 and m/z 59 significantly increased over this period, from around 0.02 ppbv s⁻¹ on DOY 85 to > 0.10 ppbv s⁻¹ on DOY 120, with a less pronounced response for m/z 45.

As the ozone input mixing ratio used during these experiments varied between 1 and 8 ppm, we normalized the measured VOC production to the ozone input. To do so, we start with the mass balance equation for the VOCs in the headspace: 1dCVOCdt=PVOCAbubVair-FairCVOCVair where CVOC represents headspace VOC mixing ratio, P(VOC) represents the net production of VOC, Vair represents headspace volume and F(air) represents gas flow rate.

As the peak height gradient (dCVOC/dt) is maximum in the first few seconds, when CVOC is still minimal, the Fair CVOC term (Eq. 1, second term RHS) is initially small and insignificant. Removing this term leads to a simplification of PVOC: 2PVOC≈VairdCVOCdtAbub The ozone (O₃) input (deposition, DO3) can be computed as: 3DO3=FairCO3,G-CO3,HAbub≈FairCO3,GAbub where CO3,G and CO3,H represent ozone input (from the generator) and in the headspace respectively. In the first few minutes of oxidation, all ozone added to the sample is reacted and so CO3,H can be assumed to be zero. The dimensionless VOC production ratio (ratio of VOC produced for a given ozone input; Fig. 2C) is calculated by dividing Eq. (2) by the approximation of Eq. (3): 4PVOC:O3=PVOCDO3≈Vair×dCVOCdtCO3,GFair Figure 3 shows the decadal averages of key oceanographic parameters at this station in weekly bins (Fig. 3a: Temperature, Fig. 3b: Salinity, Fig. 3c: Chlorophyll, Fig. 3d: Total eukaryotes, Fig. 3e: Total heterotrophic nanoflagellates and bacteria), overlaid with measurements relating to the period of this study. The site is characterised by typical temperate shelf seas dynamics, with a seasonal temperature range of 8 to 16 °C, and a large seasonal chlorophyll cycle. The L4 site is close enough to the coast to be regularly influenced by the River Tamar's freshwater plume, resulting in an average salinity of ∼ 35.0 punctuated by small dips in salinity caused by the river outflow. The period of change in VOC production from DOY 80–120 coincided with the onset of the spring bloom at station L4, with rapid rises in chlorophyll a between DOY 80–90, and a rapid rise in abundance of eukaryotes (Fig. 3c–d). This infers that the rise in abundance of phytoplankton results in the increased availability of key VOC precursors.

Following the initial spring bloom, chlorophyll a remains > 1.0 mg m⁻³ (compared to < 0.5 mg m⁻³ in winter months), with the temporal dynamics reflecting a succession of blooms of different groups of plankton, followed by a further autumn peak (DOY 260), characterised by an annual maximum in chlorophyll a concentrations (Fig. 3c–e) (Smyth et al., 2010). The variable VOC production ratios during the period DOY 165–250 is likely reflective of the impact of this highly biodynamic environment on VOC precursors. The production ratios of m/z 59 and m/z 69 were markedly variable and on average significantly higher (452 % and 155 %, respectively) than the late winter period (before DOY 80). The m/z 45 production ratio was relatively low during the early summer period (DOY 165–220) but rose by almost 300 % during later summer (DOY 220–250).

The late autumn bloom (DOY 250–280) is a period of high productivity at L4 (Fig. 3), characterised by an order of magnitude increase in production yields of m/z 69 with a mean of 95.3 ± 38.0 compared to 8.8 ± 3.5 during later winter, and less marked but significant increases in production of m/z 45 and 59 (216 % and 208 %, respectively). The highly elevated production of m/z 69 continued beyond the end of the autumn bloom, even as chlorophyll a concentration dropped back down to ∼ 0.5 mg m⁻³ (Fig. 3c). Elevated DOC concentrations in autumn have previously been observed in the nearby Celtic Sea (Carr et al., 2019), and our observations may indicate the presence of residual DOC remaining in surface waters following the period of high autumn productivity. During this latter part of the growth season, the abundance of heterotrophic nanoflagellates and bacteria also reached their highest levels (Fig. 3e), implying a regime dominated by grazing, senescence and decay, conducive to the production of DOC (Lønborg et al., 2020). Despite broadly similar temporal trends, no significant relationships were detected between VOC production ratios and concurrent chlorophyll-a or microbial abundance (r2 < 0.3), indicating that precursor composition rather than bulk biological abundance is a key control on VOC production.

The VOC production ratios presented here are normalised to total ozone input and therefore implicitly include ozone consumption by competing seawater reactants. Recent observations from a trans-Atlantic study indicate that iodide contributes, on average, approximately half of ozone reactivity in temperate waters (Yang et al., 2025). Consequently, the reported production ratios should be interpreted as an effective VOC production efficiency per mole of ozone lost from the system, rather than yields based solely on the fraction of ozone reacting with DOM. To demonstrate this, we performed experiments with the same bubble-column system (see Supplement, Fig. S3) showing that increasing iodide concentration reduced m/z 45 (acetaldehyde) and m/z 59 (acetone) production ratios, consistent with competition between iodide and DOM for available ozone.

The seasonal data is summarised in Fig. 4, allowing comparison of the production ratios for each species during each season. m/z 59 production is associated with the spring bloom and early summer – a period of vigorous phytoplankton growth. m/z 45 production peaks in the late summer, coincident with shifts in plankton community structure, and as discussed above. m/z 69 production dominates the autumn signal, showing some relationship with the abundance of grazers and bacteria. The seasonal trends we observed in VOC production imply different precursors and/or production pathways for these different compounds, which may reflect: (i) the seasonal succession of the phytoplankton community (Fig. 3c–e), (ii) the shifting abundance of organisms grazing on phytoplankton which alters DOC concentrations (e.g. heterotrophic nanoflagellates, Fig. 3e), (iii) the community structure and activity of the microbial community which may alter the availability or characteristics of VOC precursors (bacteria abundance shown in Fig. 3e), (iv) chemical processing/photooxidation of organic matter over the course of the summer months.

The suite of VOC products observed in the present experiments is generally consistent with compounds previously reported during ozonolysis of seawater and sea-surface microlayer samples. Wang et al. (2023) also observed the formation of acetaldehyde (m/z 45) and acetone/propanal (m/z 59), as well as hexanal, heptanal, octanal and nonanal during ozone-driven reactions with sea-surface microlayer samples collected from the South China Sea, while Schneider et al. (2024) reported acetaldehyde (m/z 45) and acetone/propanal (m/z 59) production from the oxidation of Arctic seawater and sea-surface microlayer samples. Kilgour et al. (2024) focused primarily on total VOC production and did not identify acetaldehyde or acetone/propanal as major products in experiments with coastal seawater. These studies did not consider the role of seasonality in influencing the influence of ozonolysis on VOC production.

Quantitative comparison between studies is complicated by substantial methodological differences. The flow-reactor studies employed by Wang et al. (2023), Schneider et al. (2024) and Kilgour et al. (2024) preferentially isolate chemistry occurring within a stable surface film, whilst the bubble-column approach generates a continuously renewed gas-water interface through bubble entrainment. The production ratios measured in the present study (acetaldehyde: 0.026–0.102; acetone/propanal: 0.023–0.124; 0.08–0.0095 for m/z 69 (see Table 1) for summary) are lower than the laboratory total VOC yields reported by Kilgour et al. (2024) (0.43–0.62) but overlap with their field-derived estimates (0.04–0.06). This closer agreement with the field-derived estimates of Kilgour et al. (2024) may indicate that bubble-column approach better captures VOC production under naturally turbulent ocean conditions. In addition, the production ratios reported here are normalised to total ozone uptake and therefore include ozone consumption by competing pathways that do not generate the VOCs measured in this study.

Our experiments above clearly demonstrate that reactions between ozone and natural seawater lead to production of VOCs, consistent with other recent work (Schneider et al., 2019, 2024; Kilgour et al., 2024). In this section, we explore potential biological VOC precursors through the addition of specific reactants to aged seawater in the experimental bubble column. Initially, additions of increasing volumes of senescent axenic algal culture (Emiliania huxleyi) showed increasing VOC peak height and production ratio (Fig. 5). Peak height of m/z 59 and m/z 45 were similar in magnitude to measurements with natural seawater, with the maximum addition of culture yielding similar peak heights to those observed during the spring bloom. This indicates that the DOM released from phytoplankton (in this case E. huxleyi) contains key VOC precursors to produce these compounds. Interestingly, m/z 69 was produced at much lower levels here relative to experiments with natural seawater, which may indicate that the full range of precursors necessary for VOC production may be dependent on natural processing of water such as bacterial metabolism of DOM, which would have been minimized under axenic culture conditions. Given the highly complex composition and spatio-temporal variability of marine DOM, a commercially certified standard is not available. We thus used a standardised Suwannee River nominal organic matter (NOM) (International Humic Substances Society) as a proxy. Additions of increasing amounts of NOM to aged seawater using a relatively high ozone input of ∼ 9 ppmv resulted in substantial ozone uptake (qualitatively similar to Martino et al., 2012) but negligible production of VOCs monitored here – evidence that unsaturated terrestrial/riverine DOM is not fully representative of VOC precursors found in surface seawater (Yang et al., 2025).