The EEC is a shallow coastal sea where depths rarely exceed 50 m, with an average depth of 45 m between Calais and Dover (Fig. 1). The region is subject to macro- and mega-tidal regimes, with tidal amplitudes often surpassing 8 m, and it receives substantial nutrient inputs from multiple estuaries along the Picardy and Opale coasts, most notably the Seine and Somme (Huguet et al., 2024). Sampling was performed weekly aboard the research vessel Sepia II (CNRS INSU-FOF, Centre National de la Recherche Scientifique Institut National des Sciences de l'Univers-Flotte Océanographique Française). The “coastal–offshore” distinction was defined within the frontal area to ensure each category consistently represented coastal and offshore waters, regardless of tidal state. Because frontal waters are strongly mixed, they span both categories, and tides mainly influence only the direction of flow (northwards or southwards; Brylinski et al., 1991). To account for tidal influences on phytoplankton communities, the sampling strategy used a high-resolution (∼ 1 km) coast–offshore transect comprising 9 stations (Fig. 1; Table 1). The transect spans approximately 9.7 km from coast to offshore, from R0 (50.8° N, 1.59° E) to R4 (50.8° N, 1.45° E) (Fig. 1). To facilitate analysis of spatial patterns, stations were categorized into two zones based on their proximity to shore: coastal (R0–R2) and offshore (R2^′–R4). The highest sampling frequency was recorded at station R1 (227 samples, Table 1). Additional details regarding data acquisition and validation are provided in Hubert et al. (2026). In addition to DYPHYRAD, we incorporated datasets from two French National Observation Systems (Systèmes Nationaux d'Observation, SNO): SOMLIT and PHYTOBS. SOMLIT contributed data from two stations (SOMLIT C and SOMLIT L) sampled at most twice per month. PHYTOBS contributed data from Boulogne and from station C, which is co-located with SOMLIT C. Data from SOMLIT C and PHYTOBS C were classified as coastal, while SOMLIT L represented offshore conditions.

At each station (R0–R4), sea temperature (T, °C) and salinity were measured using a conductivity-temperature-depth profiler (CTD Seabird SBE 25). Subsurface water samples (1–2 m depth) were collected along the DYPHYRAD transect to measure Chl-a, nutrients and phytoplankton abundance. Dissolved inorganic nutrients i.e., nitrite (NO2-), nitrate (NO3-), orthophosphate (PO43-) and orthosilicic acid (Si(OH)4) concentrations were quantified using an autoanalyzer (AutoAnalyzer ALLIANCE SpA, Italy until 2016; then AA3 HR AutoAnalyzer, SEAL Analytical GmbH, Germany) following a protocol based on Aminot and Kérouel (2004). Dissolved nutrients are referred to hereafter as DIN (NO2- + NO3-), DIP (PO43-) and DiSi (Si(OH)4). Chl-a concentrations were measured using a Turner Designs 10-AU field fluorometer (Turner Designs Ltd, USA), as described in the equations developed by Lorenzen (1966).

The DYPHYRAD dataset (Hubert et al., 2025b) comprises 1835 samples collected over 268 dates during a 10-year survey (2012–2022, Table 1). Single-cell optical analyses were conducted using pulse-shape recording flow cytometers (CytoSense and CytoSub, manufactured by CytoBuoy b.v., Woerden, the Netherlands), which are designed for high-resolution in vivo characterization of phytoplankton based on their optical properties. Four cytometers were used over the 11-year time series. These instruments count particles ranging from 0.1 to 800 µm in width, covering practically the entire phytoplankton size range. Technical specifications for each flow cytometer are detailed in previous studies (Hubert et al., 2026). Briefly, two protocols were employed: the “Pico” protocol used low detection thresholds (10 mV red fluorescence), low pump speed (5 µL s⁻¹), and short sampling time (5 min) to target 0.1–3 µm cells with low fluorescence and high abundance. The second protocol targeted nanophytoplankton and microphytoplankton using higher detection thresholds (25 mV red fluorescence), high pump speed (1013 µL s⁻¹), and longer sampling time (8–10 min). The flow cytometers are equipped with a blue laser (488 nm, 50 mW) to allow discrimination between phototrophic and non-phototrophic particles. Cytogram analysis (biplots combining scatter and fluorescence data, Hubert et al., 2025a) was performed using CytoClus 4 software (CytoBuoy b.v., the Netherlands). Five phytoplankton functional groups (PFGs) were manually discriminated and characterized based on their size distribution, structural complexity and fluorescence signals (Table 2). PHYTOBS microscopy data (https://www.phytobs.fr/en, last access: 7 September 2026) were used only to check for the species related to the observed phytoplankton blooms, while SOMLIT data (https://www.somlit.fr/, last access: 7 September 2026) provided complementary measures of temperature, salinity, nutrients (DIN, DIP and DiSi), Chl-a and flow cytometry counts of nanophytoplankton and picophytoplankton for comparison with the DYPHYRAD transect (see Fig. 1 for the locations). All samples from both networks were collected at the subsurface (1–2 m) at a bimonthly frequency. SOMLIT datasets were used solely to construct the climatologies of nutrients, salinity, temperature, and Chl-a for the study area. Chl-a was the only variable for which SOMLIT and DYPHYRAD datasets also contributed to the statistical analyses.

Storm events in this study were identified as periods when wind speeds exceeded the 90th percentile (11.3 m s⁻¹), calculated from a 25-year reference period (1 January 2000 to 30 June 2025). Storm length was identified as the ±3 day period surrounding the peak wind speed day, with day 3 designated as the storm day. Storm events were characterized using hourly measurements of wind speed, wind direction, and precipitation from the Météo-France Boulogne-sur-Mer station (50.73° N, 1.59° E; https://meteo.data.gouv.fr/, last access: 7 September 2026), located 8 km from the DYPHYRAD transect, for the period 2012–2022.

Storms were selected for analysis when phytoplankton abundance and environmental data were available for at least 10 d prior to the storm day and for 14 d following the peak storm event. Wind observations include both wind speed and direction. Wind observations were used to evaluate the wind stress at the sea surface (τ, N m⁻²) computed as (Wu, 1982): 1τ=ρairCdU2, where ρair is the air density (assumed here as ρair=1.225 kg m⁻³) and the drag coefficient Cd was assumed to depend on the wind speed as: 2Cd=0.8+0.065U10-3 The wind stress was derived from wind speed using a ±3 day window around each peak date. This approach allowed us to classify periods of low versus high wind forcing, which in turn served as an indicator of water column stability and the likelihood of wind-induced mixing.

Daily freshwater inflows were obtained from river inflow data for the Slack and Wimereux rivers, which border the transect (Fig. 1) (https://www.eaufrance.fr/, last access: 7 September 2026). River flow rates were used as indicators of riverine inputs from the respective watershed areas. For the analysis, we used the combined inflow from both rivers to estimate the mean freshwater input during a storm event. In the study area, river plumes are transported under the influence of prevailing south-westerly winds (Brylinski et al., 1991, 1996). Because of their proximity to the transect, the Wimereux Estuary and the River Slack were used as indicators of local riverine influence on water chemistry. Although the Slack lies north of the transect, strong tidal mixing, wind reversals and the NE–SW ebb-directed tidal flow (Brylinski et al., 1991) can transport its freshwater signal into the study area. Total accumulated precipitation (mm) and mean river inflow (m³ s⁻¹) were calculated over the ±3 d storm period. Thresholds for total precipitation and mean river flow were derived from the combined June–July distribution across the 25-year period, with the 80th percentiles used to define “High” and “Low” classes as indicators of storm impact.

Whisker plots were constructed to evaluate post-storm changes in Chl-a and to examine variability in Chl-a relative to the average river inflow associated with each storm. Due to the limited temporal coverage of Chl-a measurements in the DYPHYRAD dataset, Chl-a data from all available stations were used. Seasonal climatologies of phytoplankton abundances were derived from DYPHYRAD data. To assess the impact of storm events on phytoplankton abundances, we calculated the empirical complementary cumulative distribution function (CCDF, Helton, 1997), which represents the probability of observing abundance greater than or equal to a particular abundance level. In order to focus on all values, from medium to extremes, the CCDF were displayed in log-log plots to emphasize extreme events. For each phytoplankton group, the CCDF (P(X≥x)) was constructed from June–July observations from the DYPHYRAD dataset, providing a baseline distribution: 3PX≥x=1-FXx where FXx is the cumulative distribution function of a phytoplankton abundances X and x denotes a specific abundance threshold (i.e., an observed value of phytoplankton abundance).

Shaded percentile bands (50 %–90 %) were used to represent the expected June–July baseline variability, and values exceeding the 90th percentile were interpreted as extreme abundances. The pre-storm window was limited to 10 d to ensure at least one available observation per station given the weekly sampling frequency, whereas the post-storm window was extended to 14 d to capture the propagation of storm-driven freshwater and nutrient inputs into the Strait of Dover. Phytoplankton abundances from the pre-storm and post-storm periods were compared with this baseline to determine whether post-storm conditions remained within, or exceeded, the typical seasonal range. This non-parametric approach avoids assumptions about underlying distributions and is particularly suitable in this context, as phytoplankton abundances are not normally distributed (Derot et al., 2015). It also allows for a direct comparison of pre- and post-storm conditions relative to the seasonal climatology. Post-storm abundances were further examined using a distribution-wide Anderson–Darling (AD) test and a tail-exceedance test.

Based on the availability of flow cytometry data, 10 storms were retained for analysis, each showing distinct patterns in wind stress, precipitation, and river inflow. Storm-associated wind stress varied considerably among events (Fig. 2). Most storms were characterized by sustained windy conditions, with median and interquartile range (IQR) values close to 0.1 N m⁻², whereas others were weaker, marked only by short bursts of strong winds (20 June 2016, 6 June 2020, 6 June 2022). The consistently windy storms included 22 June 2012, 12 July 2012, 15 June 2013, 2 June 2015, 13 July 2015, 2 July 2016, and 27 July 2020 (Fig. 2; Table 3). Wind speeds above 8 m s⁻¹ generated wind stress exceeding 0.1 N m⁻² (Fig. 2), reaching more than 0.2 N m⁻² during extreme winds (> 11.3 m s⁻¹). The storm event on 2 July 2016 exhibited the longest duration of winds above 8 m s⁻¹ (86 h; Fig. A1 in Appendix A) and the strongest wind-stress forcing (mean = 0.12; SD = 0.09 N m⁻²; Fig. 2). Across all events, wind speeds exceeding 8 m s⁻¹ originated predominantly from south-westerly directions (Fig. A2 in Appendix A).

Precipitation amounts ranged from 1 mm (13 July 2015) to 32.20 mm (20 June 2016) over the ±3 d window surrounding each storm (Fig. 3a). High precipitation and high inflow were defined using the 80th percentile of their respective June–July distributions, corresponding to thresholds of 16.12 mm and 0.45 m³ s⁻¹ (Fig. B1 in Appendix B). In most cases, precipitation amounts corresponded with flow rates, with high precipitation events typically producing high inflows (Fig. 3b). Exceptions included the 27 July 2020 and 6 June 2022 storms, where this relationship did not hold. Similarly, the 2 July 2016 storm showed low precipitation at Boulogne-sur-Mer but was associated with elevated inflows (Fig. 3b). Across all events, inflow ranged from a minimum of 0.18 m³ s⁻¹ (13 July 2015) to a maximum of 1.98 m³ s⁻¹ (20 June 2016).

River inflow strongly influenced environmental variables (Fig. C1a–f in Appendix C). Salinity decreased below the climatological mean of 34.42 during “High” inflow storms (Fig. C1b in Appendix C). Events in 2012, 2013, and 2016 were associated with elevated inflows and a corresponding reduction in salinity (33.4–34.0; Table 3). The highest inflow rate (1.976 m³ s⁻¹ on 20 June 2016) corresponded to the lowest observed salinity (33.4). In contrast, storms with “Low” inflow (2 June 2015, 13 July 2015, 27 July 2020, 6 June 2022) had little effect on salinity (post-storm salinity > 34.0), and in the case of 6 June 2022 salinity (34.572) did not fall below the climatological mean (Table 3). Elevated inflow events were also linked to substantial increases in nutrient concentrations. Because pre-storm nutrient samples were limited (1 sample per station within the 10 d pre-storm window) and did not consistently capture conditions immediately before each event, we used climatological means as the reference baseline (Fig. C1d–f in Appendix C). Summer nutrient concentrations were typically low, while post-storm concentrations exceeding these baselines therefore indicated anomalous nutrient inputs. For example, the 20 June 2016 storm exhibited DIN and DiSi concentrations of 3.0 µmol L⁻¹, while the 22 June 2012 event, characterized by strong winds and relatively high inflow, yielded DIN concentrations of 1.52 µmol L⁻¹, DIP concentrations of 0.39 µmol L⁻¹, and DiSi concentrations of 3.136 µmol L⁻¹ (Table 3). However, due to the relatively low sampling frequency, some storm-related changes in salinity and nutrient concentrations (e.g., from mixing or riverine supply) may not have been captured, as in the cases of 2 July 2016 and 15 June 2013.

Interannual variability in the number of windy days (days with wind speed > 8 m s⁻¹; at least 10 d per month) and average river inflow (m³ s⁻¹) from the Slack and Wimereux rivers during June and July near the DYPHYRAD transect was analyzed to characterize weather patterns (Fig. 4). Pronounced interannual variability was observed in June–July wind speeds and river inflows. Years with high inflows (2012, 2013, 2016, 2021) exhibited June–July mean inflow above 0.35 m³ s⁻¹ (Fig. 4) and included storms such as 22 June 2012, 12 July 2012, 15 June 2013, 20 June 2016, and 2 July 2016. In contrast, periods with low monthly inflows (2014, 2015, 2017, June 2018, July 2019, 2020, 2022; Fig. 4) encompassed storms such as 2 June 2015, 13 July 2015, 27 July 2020, and 6 June 2022. Some years were characterised by notably windy conditions (days with wind speeds > 8 m s⁻¹; at least 10 d per month) despite low inflows, including July 2014, 2015, July 2017, July 2020, and June 2022. June–July months in 2015 and July 2020 all experienced more than 24 windy days above 8 m s⁻¹, coupled with June–July average river flow below 0.35 m³ s⁻¹ (Fig. 4). Thus, storm impacts arose either from high inflows, from strong winds alone, or from low inflows combined with short windy events (e.g., 6 June 2022). Storms were classified as consecutive when they occurred shortly after a preceding or following storm, such that the effects of the consecutive storms could have persisted. For example, the 2 July 2016 storm followed the 20 June 2016 event, suggesting that storm-driven changes in the marine environment may have been sustained between them. A similar sequence occurred in 2012, when the 12 July 2012 storm followed the 22 June 2012 event.

Storms were characterized by the biological response of Chl-a assessed through their pre- and post-storm distributions. Pre-storm mean Chl-a concentrations increased following nearly all storm events, except for those on 15 June 2013 and 13 July 2015 (Fig. 5). In high inflow storms (depicted in pink, Fig. 5), post-storm mean increases persisted above the June–July climatological mean of 3.20 µg L⁻¹. The consecutive storm events of 2012 resulted in up to a 4-fold increase in Chl-a concentrations, whereas the consecutive events of 2016 maintained concentrations above the 3.20 µg L⁻¹. A similar persistence of elevated Chl-a was observed following the 15 June 2013 storm, despite the relatively low inflow (depicted in blue, Fig. 5) of 0.35 m³ s⁻¹. A subsequent event on 22 June 2013, with a mean flow rate of 0.39 m³ s⁻¹, occurred within the 14 d post-storm window and is therefore considered part of the same response as the 15 June 2013 storm. The remaining low inflow storms (depicted in blue, Fig. 5) did not elevate post-storm mean Chl-a concentrations above 3.20 µg L⁻¹. Post-storm responses in Chl-a distribution were more closely aligned with river inflow than with precipitation, as illustrated by the 27 July 2020 and 6 June 2022 storms (Fig. 5; Fig. E1 in Appendix E).

The 15 d climatology derived from weekly to biweekly data captured spatial and temporal variations in phytoplankton abundances within each month and revealed that distinct groups characterize different periods of the year (Fig. 6a–f). RedMicro (diatoms) reached abundances on the order of 10⁵ cells L⁻¹, increasing from late winter to late spring and declining during summer (Fig. 6a). A sharp increase in RedNano (10⁷ cells L⁻¹) from late March to early May reflected the recurring Phaeocystis globosa blooms in EEC waters (Fig. 6b). After spring, RedNano abundances were sustained at ∼ 10⁶ cells L⁻¹. Other nanophytoplankton groups showed distinct seasonal peaks: HsNano (coccolithophorids) in summer and autumn (Fig. 6c), and OraNano (cryptophytes), which increased steeply from June through autumn, reaching ∼ 10⁵ cells L⁻¹ (Fig. 6d). During summer, the phytoplankton community was dominated by picophytoplankton, with abundances on the order of 10⁷ cells L⁻¹ (Fig. 6e–f), including OraPicoProk (Synechococcus) and RedPico (picoeukaryotes). The climatology also revealed high standard errors in diatom, picophytoplankton, and cryptophyte abundances, indicating greater variability during June–July. Spatially, RedMicro were more abundant in coastal waters (Fig. 6a), whereas Synechococcus dominated offshore waters (Fig. 6e).

An empirical complementary cumulative distribution function (CCDF) was applied to detect shifts in phytoplankton abundances by quantifying the probability of blooms exceeding climatological ranges (Fig. 7A–D). Four storm types were considered. The first type was characterized by high inflow with low wind stress, exemplified by the storm of 20 June 2016, which was the rainiest and had the highest inflow, while showing the least impact of wind. The second type consisted of consecutive storms, with the storm of 2 July 2016 occurring immediately after the previous storm on 20 June 2016. The third type was defined by low inflow with high wind stress, represented by the storm of 13 July 2015, which had the lowest precipitation and inflow but was marked by strong winds. The fourth type involved low inflow with low wind stress, illustrated by the short storm of 6 June 2022, which was classified as high in precipitation yet remained among the lowest in inflow and wind stress. The CCDFs of the remaining six storms are presented in the appendices with these storm characterizations (Figs. F1–F6 in Appendix F).

In During June–July, south-westerly winds frequently prevail, often exceeding 8 m s⁻¹ (Figs. A1–A2 in Appendix A). These winds enhance vertical mixing and drive the northward transport of deep and oceanic waters, replenishing nutrients in surface layers (Dulière et al., 2019). In addition, high precipitation events can produce either high or low inflow (Fig. 3). At the same time, south-westerly winds sustain a narrow coastal current, 3–4 km wide, originating from the Bay of Somme and occasionally linked to the Seine Estuary (Brylinski et al., 1991). This current is distinct from offshore waters, as reflected in the coastal–offshore salinity gradient, and serves as a conduit for riverine inputs from the Somme, Authie, Canche, Liane, Wimereux and Slack rivers. Although we did not identify a measurable effect of tidal phase on the post-storm observations (not shown here), previous studies indicate that tides primarily influence biological processes through their control of the width of the coastal current and associated transport of riverine inputs, rather than through the coastal–offshore tidal cycle itself (Brylinski et al., 1996; Fettweis et al., 2022). Storm impacts are reflected in the coastal water salinity climatology, which exhibits an increasing trend until late May, followed by a decline through the end of July (Fig. C1b in Appendix C). Under south-westerly winds (Figs. A2 and D1 in Appendix A and D), these surplus inputs can be advected northward into the study area, ultimately reaching the Strait of Dover (Brylinski et al., 1996). For example, the Somme Estuary delivers daily averages of ∼ 500 kmol d⁻¹ of silicate and nitrate during June–July (Loquet et al., 2000). Such enrichment from the Bay of Somme has long been associated with the proliferation of diatoms and Phaeocystis globosa in the EEC (Brunet et al., 1996).

In the English Channel and southern North Sea, summer is often associated with storm activity (Cook et al., 2015; Gronholz et al., 2017; Scholz et al., 2022). Extratropical storms generate extensive cloud cover, precipitation, and riverine inflows, all of which can alter light availability, nutrient concentrations and salinity (Jennings et al., 2012; Kasprzak et al., 2017). The EEC summer months are typically marked by low silicate, nitrate, and phosphate concentrations (Fig. C1d–f in Appendix C). Chl-a concentrations were relatively high and showed spatial differences, with higher values in coastal waters compared with offshore waters during the summer months (Fig. C1c in Appendix C). As previously reported, phytoplankton blooms in the English Channel can persist for several weeks after summer storms (Rees et al., 2009). For instance, Rees et al. (2009) documented three summer precipitation events that induced freshwater stratification and delivered DIN to surface waters, stimulating primary production. They proposed that storm-driven nutrient loading alleviated DIN limitation, while dissolved organic phosphate was mobilized to meet DIP demand. Such processes may generate synergistic, non-additive responses to dual nutrient enrichments, resulting in enhanced phytoplankton growth (Allgeier et al., 2011).

Interannual variability in windy days (number of days > 8 m s⁻¹) and inflow during June–July was evident, with episodes of strong winds in 2015 and 2020, and enhanced riverine inflow in 2012, 2013, 2016, and 2021 (Fig. 4). Long-term records show that northern Europe has become wetter since the last century (Cook et al., 2015), with alternating wet and dry periods driven by both climate change and natural variability. Corresponding variability was reflected in post-storm Chl-a concentrations and phytoplankton community structure. Since 2012, Chl-a in the DYPHYRAD transect has declined (Hubert et al., 2025a), consistent with earlier observations of interannual shifts in the EEC, including a late-20th-century decrease followed by a rise in the early 2000s (Lefebvre et al., 2011). Similar patterns have been reported elsewhere, with summer primary production responding to precipitation and riverine inputs in the western English Channel (Barnes et al., 2015) and to increased summer winds in the Arctic Ocean (Crawford et al., 2020). These findings suggest that interannual and multidecadal variability in Chl-a may be in part also regulated by distinct storm impacts affecting phytoplankton composition (Stockwell et al., 2020).

Previous studies have shown that large phytoplankton often dominate nutrient-rich upwelling regions of the ocean, whereas open-ocean waters are typically dominated by picophytoplankton (James et al., 2022; Ribalet et al., 2010). A similar pattern was observed in the English Channel, where riverine input represents the dominant mechanism of nutrient supply. This nutrient enrichment supports the abundance of larger phytoplankton, such as diatoms and Phaeocystis globosa, in coastal waters (Napoléon et al., 2014). During the summer months, distinct and contrasting patterns emerged between coastal and offshore waters: diatoms were consistently more abundant in coastal waters (Fig. 6a), whereas Synechococcus dominated offshore waters (Fig. 6e). Seasonal dynamics revealed a June–July mean increase in diatom abundance following the spring bloom. Synechococcus rose in late May but dipped in early July despite warming temperatures, then recovered by late July as diatoms declined. This crossover suggests that transient meteorological disturbances, including storm-related reductions in light or shifts in nutrient dynamics, temporarily favoured diatoms over Synechococcus. The subsequent reversal toward late July reflects a return to more stable summer conditions. Similarly, high standard errors in diatom, picophytoplankton, and cryptophyte abundances point to greater fluctuations during June–July (Fig. 6a–f) due to storm-driven disturbances.