ERA5, the fifth-generation reanalysis for the global climate and weather produced by the European Centre for Medium-Range Weather Forecasts (ECMWF), provides high-resolution, consistent, and quality-controlled datasets extending back to 1940 (product ref. no. 1 in Table 1). ERA5 offers hourly estimates of a wide range of atmospheric and oceanic variables, which are regridded, respectively, to a regular 0.25° and 0.5° horizontal grid. Synoptic-scale analyses of daily-averaged fields of sea level pressure (SLP), 10 m wind speed (W10), significant wave height (SWH), and mean wave direction (MWD) were conducted over the North Atlantic domain (40° W–20° E, 20–70° N) to characterize the dominant metocean features during the peak phase of Storm Louis (Table 2, no. 1).

Two deep-water and two coastal moored buoys, operated by Puertos del Estado, were utilized in this study (Figs. 1d–e and A1). These in situ observational platforms provide quality-controlled, hourly-averaged measurements of a range of oceanographic parameters (product ref. no. 2 in Table 1), including SWH, MWD, wave period at spectral peak (or peak wave period, PWP), and mean wave period (MWP) (Table 2, no. 2 and no. 3). Additionally, deep-water moored buoys also provide hourly-averaged estimations of atmospheric parameters (SLP and wind). Comprehensive technical specifications for each buoy are detailed in Table 3.

The quality control, defined by the Copernicus Marine Service in situ team (Copernicus Marine In situ Team, 2020), was based on a battery of automatic checks performed in real time to flag and subsequently filter inconsistent values: the spike test, stuck value check, and rate of change over time, among others.

Two radar tide gauges, manufactured by Miros and operated by Puertos del Estado as part of the REDMAR network (Pérez et al., 2014), have been deployed at the mouth of Langosteira (Figs. 1d and A1a) and Tarragona (Figs. 1e and A1b) harbours since November 2012 and May 2011, respectively (Table 4). These sensors deliver quality-controlled sea level measurements at a sampling rate of 2 Hz (product ref. no. 3, Table 1), facilitating the analysis of high-frequency oscillations such as meteotsunamis and infragravity waves (García-Valdecasas et al., 2021). Complementarily, 20 min averaged estimates of port agitation – defined as port-scale oscillations induced by incoming wind waves with periods shorter than 30 s – were derived using an eighth-order Butterworth high-pass digital filter with a cut-off frequency of 1/30 Hz. The resulting agitation time series were subsampled to hourly intervals (Table 2, no. 3) for subsequent analysis: i.

The agitation time series were analysed over the period of 20–29 February 2024 to evaluate harbour response under extreme weather conditions. Complementarily, hourly storm surge estimates (i.e. meteorological residuals) were examined to assess their potential contribution to wave-induced agitation during Storm Louis.

Two “direction-finding” SeaSonde HFR networks (Barrick et al., 1977), manufactured by CODAR Ocean Sensors, were employed in this study (Table 5). While the first one (composed of five sites) was deployed along the northwestern Iberian Peninsula (HFR-Galicia) in July 2010, the second one (formed by three sites) was installed in the northeastern Iberian Peninsula (HFR-DeltaEbro) in December 2013. The geographical distribution of the radar stations is shown in Fig. 1d and e, respectively.

In both study regions, the HFR installations provide wide and unobstructed coverage, ensuring stable and reliable directional solutions. In the case of the HFR-Galicia network, all HFR sites are installed at lighthouse locations, which offer elevated, exposed positions with a clear line of sight over the ocean and minimal physical obstructions. In the HFR-DeltaEbro network, the HFR sites are deployed within a protected natural area, where infrastructure is limited and the surrounding environment is well preserved, resulting in minimal electromagnetic or physical interference. These site characteristics ensure favourable geometric configurations and stable directional performance for wave retrieval in both regions.

Each network provides half-hourly wave observations (product ref. no. 4, Table 1) – including SWH, centroid wave period (CWP), and MWD – retrieved in near real time across multiple annular rings (circular concentric range arcs) extending radially from each onshore radar site (Fig. A2). For this study, HFR-derived wave measurements were subsampled to 60 min intervals (Table 2, no. 4) to ensure temporal consistency across datasets, enabling not only the investigation of wave conditions during Storm Louis but also a robust long-term skill assessment over the 2014–2024 period.

To elucidate the driving mechanisms underlying the development of Storm Louis, the prevailing atmospheric conditions were briefly examined at the synoptic scale, with particular emphasis on daily mean fields of SLP and W10. This analysis aimed to identify the large-scale barometric patterns, pressure gradients, and wind structures that contributed to the storm's intensification, thereby enabling a better understanding of the meteorological forcing responsible for the observed oceanographic impacts, especially within the affected harbour environments.

Recognizing the interconnected nature of hydrodynamic variables, an isolated analysis of individual extremes was deemed inadequate to capture the full spatial and temporal footprint of Storm Louis. Instead, a percentile-based statistical framework was employed to characterize the tail behaviour of key environmental variables, including W10, SWH, and MWP, among others. The 1st, 5th, 25th, 50th, 75th, 90th, 95th, 99th, and P99.9th historical percentiles (P1, P5, P25, P50, P75, P90, P95, P99, and P99.9, hereinafter) were derived from an 11-year (2014–2024) observational dataset to contextualize storm Louis within its broader climatological distribution (Coles, 2001). These high-impact thresholds may serve as operational benchmarks for impact-based analysis, facilitating the identification of critical exceedance frequencies relevant to risk assessment and port operational management. Notably, similar percentile-based thresholds have previously been proposed in the context of coastal storm characterization by Harley (2017), Morales-Márquez et al. (2020), and Fanti et al. (2023). In the specific case of port agitation, historical percentiles were intended as purely statistical indicators of anomalous harbour response rather than direct representations of formal operational limits applied by port authorities. Hourly estimations of storm surge components (meteorological residuals) were analysed to assess their potential contribution to the pronounced increase in harbour agitation observed during Storm Louis. This evaluation aimed to determine whether elevated water levels played a significant role in amplifying the port response under storm conditions. In this context, connected extremes are of particular concern for harbour operability. Their combined impacts can interact nonlinearly, resulting in synergistic effects that pose a greater threat to port infrastructure than isolated extreme events (Velpuri et al., 2023).

The long-term relationship between energetic offshore wave conditions (SWH, MWP, and MWD) and port agitation was analysed to assess directional dependencies and harbour response under extreme-forcing scenarios, as the direction of the most energetic incoming waves may be critical to harbour impacts. Scatter plots were thereby computed to infer the dependence between port agitation and offshore SWH data.

In addition, the monthly distribution of extreme hourly agitation events – defined as values exceeding the P90, P95, and P99 thresholds – was derived from the 11-year observational time series provided by the Langosteira (NWI) and Tarragona (NEI) tide gauges. This analysis aimed to identify potential seasonal patterns or preferential timing in the occurrence of high-agitation conditions, thereby providing insight into whether certain phases of the annual cycle are more prone to operational disruptions due to extreme metocean forcing.

The HFR land-based remote sensing technology is based on measurements of the radio-wave-backscattered signal from ocean surface gravity waves in the 3–30 MHz range of the electromagnetic spectrum (Crombie, 1955). CODAR SeaSonde “direction-finding” HFR networks measure the return signal continuously over all angles, exploiting the directional properties of a collocated three-element receive antenna system to infer the direction of the incoming signals (Barrick et al., 1977).

While fine-resolution maps of the surface circulation are obtained from the Doppler shift of the predominant first-order peaks in the sea-echo Doppler spectrum, wave parameters are deduced from the weaker and noisier second-order peaks. The second-order scattering-based methods rely significantly on echo quality, which varies with sea state and radar frequency (Wyatt et al., 2005). The relative contribution of the second-order spectrum increases with both the radar frequency and the wave height since wave data are dependent upon the occurrence of both Bragg and larger surface gravity waves. There is a minimum threshold for sea states at a given radar frequency in which reliable wave parameters can be determined. Below such a sensitivity threshold, the lower-energy second-order spectrum is closer to the noise floor and more likely to be contaminated with spurious contributions that might result in wave height overestimation or limited temporal continuity in wave measurements (Tian et al., 2017; Lipa et al., 2022; Lipa and Nyden, 2005). During extreme weather events, there is also a limiting factor for HFR accuracy as the wave height increases and exceeds the saturation limit defined (on an inverse proportion) by the radar transmit frequency. If the radar spectrum saturates, the first- and second-order peaks merge, and interpretation of the spectra becomes impossible with existing methods (Forney et al., 2015). It is worth mentioning that theoretical upper performance thresholds for each HFR system (Table 5) were not exceeded during Storm Louis.

The HFR directional wave spectrum and derived parameters (SWH, CWP, and MWD) can be determined from the weaker second-order sea-echo Doppler spectrum by adopting two main approaches: full integral inversion or fitting with a model of ocean wave spectrum (Lipa and Nyden, 2005). A variety of inverse techniques have been developed over the last few decades (Barrick et al., 1977; Wyatt, 1990; Hisaki, 1996). The wave parameters used in this study, based on 30 min averaged backscatter data, were obtained directly by CODAR radar proprietary software: SeaSonde Radial Suite Software Release 7 (R7). This wave-processing tool performs a least squares fitting technique between the second-order radar spectrum and a Pierson–Moskowitz with cardioid directional function model. Specifically, SeaSonde HFR networks measure a wave period that represents the centroid of the model being fitted to the second-order Doppler spectrum (Lipa and Nyden, 2005), hence hereafter referred to as the centroid wave period (CWP).

HFR wave data were collected in near real time, assuming homogeneity over the entirety of each range cell (RC). Specifically, each site including the HFR-Galicia system has five individual RCs (5.1 km wide), which extend radially from an origin at the onshore radar site to a distance 25.5 km offshore (Fig. A2a). Each site from HFR-DeltaEbro presents seven RCs (1.66 km wide), which extend radially from an origin at the onshore radar site to a distance 11.62 km offshore (Fig. A2b). While the returned signal might be too weak for the outermost RCs, shallow-water effects can become significant for innermost RCs, also impacting on radar sea echo (Lipa et al., 2008). Therefore, we selected the intermediate RC that recorded the highest percentage of valid data, based on the same quality control procedures previously applied to the in situ wave observations. Accordingly, RC3 from the VILA site (HFR-Galicia) and RC4 from the ALFA site (HFR-DeltaEbro) were selected for comparison with in situ wave observations from Villano-Sisargas and Tarragona moored buoys, respectively, in order to evaluate the HFRs' accuracy in capturing extreme offshore wave conditions (Fig. A2). The selected sites were those closest to their corresponding deep-water moored buoys and/or those that exhibited the highest wave data availability throughout the lifecycle of Storm Louis.

Finally, the statistical metrics used in the present study to compare two datasets included the mean, the standard deviation (σ), the root mean squared error (RMSE), and the Pearson correlation coefficient (R) (Emery and Thompson, 2001).

The large-scale synoptic conditions associated with Storm Louis were characterized using daily-averaged ERA5 reanalysis data for SLP, W10, and SWH (Fig. 1a–c). The prevailing SLP pattern featured a classic dipole structure, consisting of a persistent subtropical high (∼ 1035 hPa) over the southwestern North Atlantic and a deep low (∼ 968 hPa) centred over northeastern Europe (Fig. 1a). This configuration, commonly referred to as the Azores High–Icelandic Low dipole, is a key expression of the North Atlantic Oscillation, a dominant mode of atmospheric variability modulating storm tracks and weather extremes across the North Atlantic sector (Hurrell and Deser, 2009). During the peak phase of the storm (23 February 2024), the pressure gradient exceeded 1.7 Pa km⁻¹ and was oriented along a southwest–northeast axis, promoting sustained northwesterly winds with velocities surpassing 18 m s⁻¹ over the northeastern Atlantic (Fig. 1b). These intense winds generated high-energy swell systems with SWH exceeding 8 m (Fig. 1c).

To place the magnitude of Storm Louis in context, these values can be compared with those reported for the exceptional Storm Gloria in January 2020 (Álvarez-Fanjul et al., 2022). During Gloria, wind speeds also exceeded 18 m s⁻¹, while SWH and MWP ranged between 4.5–7.6 m and 7–9 s, respectively (Lorente et al., 2021). The similarity in wind forcing and SWH values suggests that Storm Louis may be considered an event of comparable meteo-oceanic impact.

In particular, the NWI study area (Fig. 1d) and Langosteira harbour (Fig. A1a) experienced substantial wave impact and atmospheric pressure anomalies. The Villano-Sisargas deep-water buoy recorded a cumulative SLP drop of 33 hPa between 20–25 February, reaching a minimum of 1000 hPa (Fig. 2a), a value slightly below the P5 threshold derived from the 2014–2024 reference period (Fig. 2b). Concurrently, peak offshore wind speeds reached 17.57 m s⁻¹, exceeding the local P99.9 percentile (Fig. 2c). The Langosteira coastal buoy detected extremely energetic sea states from the northwest (Fig. 1d), with a peak SWH of 8.44 m and a peak MWP of 13.7 s recorded on 24 February (Fig. 2a). Notably, the P99.9 threshold of SWH (8.04 m) was occasionally exceeded during 2 non-consecutive hours, while the observed MWP moderately surpassed the P99.9 level (12.0 s), nearly matching the all-time maximum value of 13.9 s recorded during the 2014–2024 reference period (Fig. 2d).

Severe oceanographic conditions were also observed in the NEI study area (Fig. 1e). Near Tarragona harbour (Fig. A1b), the deep-water buoy recorded an SLP minimum of 999 hPa (Fig. 2e), approaching the P1 threshold for the region (Fig. 2f). Offshore wind speeds peaked at 14.29 m s⁻¹, exceeding the local P99 (Fig. 2g). The Tarragona coastal buoy recorded high waves coming from the southernmost sector (Fig. 1e), with SWH and MWP peaking at 1.67 m and 6.04 s, respectively, during the early hours of 26 February (Fig. 2e). These values were close to or marginally above the P99 thresholds for the corresponding wave parameters (Fig. 2h).

Hourly time series of the storm surge component (meteorological residuals) and port wave agitation, measured during the full lifecycle of Storm Louis at the Langosteira (NWI area) and Tarragona (NEI area) tide gauges, are presented in Fig. 3a and b, respectively. While both sites experienced high wave agitation above historical P99 thresholds, the intensity and temporal structure of these events differed substantially. The Langosteira tide gauge displayed a moderate but prolonged agitation episode, reaching a peak of 0.68 m on the afternoon of 24 February (Fig. 3a). The P99 threshold for the 2014–2024 period (set to 0.52 m) was surpassed slightly for 33 non-consecutive hours. The storm surge component remained relatively subdued throughout the event, with fluctuations mostly below 0.50 m and no clear peak coinciding with the maximum agitation. By contrast, the Tarragona tide gauge captured an intense and short-lived event, peaking at 1.77 m early on 26 February 2024 (Fig. 3b). The local P99 threshold for the 2014–2024 period (set to 1.20 m) was far exceeded for 16 consecutive hours. The storm surge signal at the Tarragona tide gauge remained modest, with magnitudes generally under 0.3 m, exhibiting minimal correlation with the agitation peak, consistent with a wave-driven dynamic.

These contrasting behaviours can be interpreted within a multi-scale framework linking large-scale wave generation to local harbour response. Wave energy generated in the North Atlantic required a finite amount of time to propagate toward the regional offshore domains and underwent directional evolution before interacting with each harbour. At the local scale, harbour agitation did not respond instantaneously to offshore wave forcing; rather, it was strongly modulated by site-specific factors, including directional sensitivity, wave transformation processes at the harbour entrance, and intrinsic response times associated with wave–structure interactions. Processes such as diffraction and reflection within the harbour basins played a key role in shaping both the magnitude and the temporal persistence of the observed agitation, thereby explaining the pronounced differences between the two sites despite their exposure to the same synoptic-scale storm event.

Hourly scatter plot analyses conducted over the 11-year study period (2014–2024) revealed a strong statistical correlation between offshore wave conditions and internal harbour agitation across both study sites (Fig. 3c–f). Linear regression analysis of the SWH against harbour agitation yielded significantly high correlation coefficients, specifically R = 0.88 for Langosteira harbour and R = 0.75 for Tarragona harbour. In the case of Langosteira harbour (Fig. 3c–d), a total of 743 hourly records exceeded the P99 agitation threshold (0.52 m). Of these, 100 % were associated with MWP above 6 s and incident wave directions within the prevailing sector of 295–360° (clockwise from true north). By contrast, the scatter plot for Tarragona harbour exhibited a more complex, bimodal directional response (Fig. 3e–f). Despite the highest offshore SWH (> 3 m) and MWP (> 7 s) values being observed from wave directions between 90 and 120°, these did not correspond to peak agitation events within the port basin. Instead, agitation events exceeding 2 m were predominantly driven by moderate offshore wave heights (∼ 2–3 m), associated with MWP values in the range 5–7 s, penetrating from the southernmost sector (150–210°). This indicates a pronounced sensitivity of the harbour's internal hydrodynamics to wave approach angles, likely modulated by local bathymetric features and structural configurations at the port entrance (Fig. 3c and e, respectively).

In addition, the monthly distribution of extreme-wave-induced agitation was assessed over an 11-year period (2014–2024), as shown in Fig. 3g–h. A pronounced seasonal signal was evident in Langosteira harbour, with peak exceedance frequencies occurring during the boreal winter months (Fig. 3g). Specifically, January through March exhibit the most severe agitation conditions: cumulative exceedance percentages over the 3-month period reached approximately 57 %, 61 %, and 68 % of total hours above the P90, P95, and P99 thresholds, respectively. As the annual cycle advances, a marked decline in exceedance frequency is observed, reaching near-negligible levels between June and August. This attenuation coincides with the seasonal minimum in offshore wave energy flux. From September onwards, there is a gradual resurgence in the frequency of agitation exceedance, with a secondary peak observed in October through December. This bimodal distribution suggests that, in addition to the primary forcing by winter storm activity, transitional swell events and associated synoptic-scale disturbances during the late autumn months contribute substantially to Langosteira harbour agitation.

By contrast, the monthly distribution at Tarragona harbour revealed a more evenly spread regime for extreme agitation (Fig. 3h). High exceedance percentages were evident during the first 4 months of the year, particularly in March, where the percentages across all thresholds approached 16 %–17 %. Unlike Langosteira harbour, where agitation was heavily concentrated in January and February, Tarragona harbour exhibited a more gradual onset and decay of agitation activity during the winter-to-spring transition. From June to September, the proportion of hourly agitation exceeding any threshold fell below 5 %, marking a seasonal minimum consistent with the calm summer wave climate typical of the western Mediterranean Sea. A second rise in extreme agitation was observed in the final quarter of the year. November and December, in particular, exhibited a high proportion of hours exceeding the P99 threshold, reaching nearly 15 %. These outcomes underscore Tarragona harbour's sensitivity to storm events, particularly those originating from southern or southeastern sectors, which are known to be energetically effective in exciting agitation in this specific port geometry.

A quantitative skill assessment of HFR-derived wave estimates during Storm Louis was performed. In the NWI region, time series from the VILA site and the Villano-Sisargas deep-water buoy exhibited strong visual agreement throughout the extreme event (Fig. 4a). The correlation coefficient for SWH is high (R = 0.92), with an RMSE of 0.96 m, indicating a high degree of coherence. Both sensors captured the initial wave growth from 21 February, the primary energy peak between 24–25 February, and a secondary crest around 26–27 February. The VILA site slightly overestimated SWH maxima relative to the buoy, particularly during the storm's mature phase, although both records converge during the storm's decay stage. The concurrent wave period time series in the NWI area showed moderate agreement (R = 0.66, Fig. 4b), with the HFR-derived CWP and buoy PWP estimates both resolving the rise to values above 14 s during the 21–22 February surge, a brief drop around 23 February, and a sustained 13–15 s regime during the storm's main phase. The HFR data tend to smooth out high-frequency fluctuations, leading to an underestimation (∼ 1 s) of the maximum buoy-reported wave periods (16–17 s) and an overestimation during the waning stage of the storm. It is worth mentioning that SeaSonde HFR networks measure a wave period (CWP) that represents the centroid of the model being fitted to the second-order Doppler spectrum (Lipa and Nyden, 2005). Thus, HFR centroid periods always dropped between PWP and MWP collected at the Villano-Sisargas buoy. This fact is relevant to properly interpret the skill metrics obtained from comparison, which was conducted against PWP.

By contrast, wave field dynamics in the NEI sector exhibited lower overall energy levels. Both the ALFA site and the Tarragona deep-water buoy recorded relatively calm conditions until 23 February, followed by a marked intensification and a first peak on 26 February. After a rapid decline, a new peak reaching 2.6 m was observed on 28 February (Fig. 4c). Agreement is weaker in this region, with a correlation coefficient of 0.60 and an RMSE of 0.51 m. Wave period estimates in the NEI region (Fig. 4d) demonstrated poorer agreement (R = 0.34 and RMSE = 1.55 s). The buoy data exhibited pronounced high-frequency variability, with wave periods ranging from 3 to 10 s. However, the HFR-derived CWP was smoother, capturing the dominant 5–8 s wave band during 24–26 February but underrepresenting both short-period (< 4 s) and long-period (> 9 s) oscillations by up to 2 s.

Overall, the results confirm that HFR networks effectively captured the general evolution and magnitude of extreme-storm-induced wave fields, exhibiting consistently higher skill in the estimation of SWH compared to wave period. Performance was notably better in the NWI region, likely reflecting the combination of higher incident wave energy and more comprehensive radar coverage. This spatially dependent variability in performance is consistent with the results reported by Saviano et al. (2022), who documented similar HFR–buoy skill contrasts under differing wave energy regimes and bathymetric conditions. These findings underscore both the potential and the limitations of HFR technology for real-time wave monitoring under severe weather conditions, particularly in high-energy offshore environments.

Additionally, a comprehensive long-term validation of HFR-derived wave parameters provided by the VILA site was conducted using in situ measurements from the Villano-Sisargas deep-water buoy over an 11-year concurrent period (2014–2024). The scatter analysis of SWH (Fig. A3a) demonstrated strong agreement between the two sensors, with a significantly high correlation coefficient (R = 0.87), a moderately low RMSE of 0.82 m, and a best-fit regression slope near unity (1.02). The negligible intercept (-0.01 m) and close clustering along the 1 : 1 reference line further confirm the HFR system's ability to capture both the magnitude and the variability of SWH over extended timescales. By contrast, the comparison of wave periods (Fig. A3b) yielded a moderate yet statistically significant correlation coefficient (R = 0.71) accompanied by a higher RMSE of 1.61 s. The best-fit slope of 0.61 and intercept of 4.25 s indicate a systematic divergence between the HFR-derived CWP and the buoy-derived PWP. Specifically, the HFR tends to underestimate high-period events, while overestimating lower-period regimes, reflecting inherent methodological differences in how each system resolves wave spectral characteristics. For MWD (Fig. A3c), the scatter distribution is more dispersed, with a reduced slope (0.52), an intercept of 135.28°, and an RMSE of 33.97°. Despite a moderate correlation (R = 0.66), the observed bias and compression of the directional range suggest limitations in the angular resolution of the HFR system, particularly at longer ranges where reduced Bragg scattering coherence may degrade directional accuracy.

These outcomes, which are in accordance with previous statistical results reported in the literature (Atan et al., 2015; Long et al., 2011), collectively demonstrate that the VILA site provided robust long-term estimates of SWH, moderately reliable wave period data, and qualitatively accurate but less precise wave direction estimates. The performance differences across parameters reflect known limitations of HFR networks in estimating wave direction and longer-period waves, especially under complex sea states or reduced HFR return conditions. Although the unimodal Pierson–Moskowitz spectral model has proven suitable for wind- and swell-dominated conditions, its applicability may be limited in complex metocean environments characterized by multimodal sea states (Lipa et al., 2022). It is also important to note that the substantial separation between the deep-water buoy locations and the selected RCs – 22.52 km for HFR-Galicia (Fig. A2a) and 47.39 km for HFR-DeltaEbro (Fig. A2b) – constitutes a potential source of discrepancy that may affect the derived skill metrics and partially accounts for the observed differences in performance between the two study areas. Moreover, the assumption of spatial homogeneity within each circular RC in CODAR SeaSonde HFR systems may further contribute to the observed discrepancies, particularly in intricate coastal environments where pronounced bathymetric variability and wave transformation processes – such as refraction, dissipation, and breaking – play a dominant role (Wyatt et al., 2005).