Sea level variations are affected by a combination of various physical processes encompassing a broad spectrum of time scales. These processes, typically distinguished by their period, wavelengths, amplitudes, repeatability, and manifestation, include short-term phenomena like surface gravity waves lasting from 1 to 20 s, infragravity waves, long ocean waves, seiches, and tsunamis with periods ranging from minutes to over 1 h, tidal oscillations peaking at ∼1/2 and ∼1 d, and other sea level variations which occur due to meteorological, hydrological and circulation effects that last from several hours to years (including interannual and decadal fluctuations), as well as long-term changes of mean sea level due to planetary, geological and climatological factors (IOC, 2006; Pugh and Woodworth, 2014). To accurately distinguish and analyse these processes, it is essential to measure sea level using sampling intervals and instruments appropriately suited to the specific temporal and spatial scales of interest.

The first sea level measurements were conducted in 1679 at Brest, France (Picard and de la Hire, 1680) and later at various locations across northern Europe such as Amsterdam in 1682 (van Veen, 1954), Liverpool in 1768 (Woodworth, 1999) and Stockholm in 1774 (Ekman, 1988), using either visual observations of graduated markings on stone walls of docks (Woodworth et al., 2015) or wooden graduated rods, called tide poles, with both of these located at points where observers could read off the instantaneous sea level height at any given time (Woppelmann and Pirazzoli, 2019). Initially, sea level measurements were conducted with sampling intervals between one reading per month and one per week. In the mid-19th century, the sampling interval of these measurements increased to 1–2 times per day, focusing specifically on high and low water levels and their respective timings (Cartwright, 1999). Naturally, sea level variations occurring at periods shorter than 2–4 d were not distinguishable by such measurements.

The development of the first automatic or “self-registering” tide gauge in the UK around 1830 (Palmer, 1831) marked a significant milestone in sea level measurement history, enabling the production of continuous sea level traces for the first time. These traces were, at best, digitized with an hourly resolution allowing for the analysis of tides, storm surges, and long term mean sea level changes. However, oscillations occurring at periods shorter than 2 h, including wind waves, infragravity waves, tsunamis, and other long ocean waves, could not be discerned from hourly values (IOC, 2006). To assess these shorter period oscillations, one had to digitize sea level traces with a higher resolution – a strenuous procedure which was only sporadically applied on episodes of high interest (e.g., Šepić et al., 2012). The wind waves, though, could not be obtained by this procedure, as they are filtered out from the sea level records during the measurement phase. This is because traditional float tide gauges typically measure sea level in stilling wells in which oscillations of the shortest periods are damped. Modern sea level measuring instruments (such as radar gauges) also usually do not measure wind waves due to averaging or filtering procedures implemented in the instrument's software.

For more than two decades, various new technologies have been developed to accurately and reliably measure sea levels with 1 min and even shorter time steps. These technologies include floats connected to digital converters, bottom pressure recorders, and acoustic and radar systems (IOC, 2006, 2016). Nonetheless, measuring sea level with an hourly time step endured as a standard approach in most countries, in accordance with the GLOSS program's initial requirements (IOC, 1990, 1997), at least until the catastrophic Indian ocean tsunami in 2004 (Lay et al., 2005). In response to this devastating event, the Intergovernmental Oceanographic Commission (IOC) of UNESCO took swift action and established multiple Intergovernmental Coordination Groups to implement effective tsunami warning systems and advised the installation of new sensors with a higher sampling frequency (1 min or less). In addition to greatly improving tsunami warning services, it also enabled enhanced research of high-frequency sea level oscillations, encompassing tsunamis (including meteotsunamis), infragravity waves, and seiches (IOC, 2006). The upgrade was facilitated by advancements in technology, particularly the further improvement and development of measuring sensors and communication links (Pérez et al., 2013).

Nowadays, sea level data can be found in various databases. These include global, regional and national databases, some of which are freely available, while others are accessible under specific conditions. Unfortunately, certain databases are unavailable due to data ownership restrictions, which strictly limit data exchange (e.g., indicated in Pérez Gómez et al., 2022 for the Mediterranean). The following data centres and databases contain freely available global sea level data acquired from tide gauge observations:

Permanent Service for Mean Sea Level (PSMSL; https://www.psmsl.org/, last access: 15 June 2026) collects and publishes monthly and annual mean sea level values from the global network of tide gauges operated by local and national authorities (Holgate et al., 2013);

Among the databases and datasets mentioned, as it collects data in real time, the IOC database stands out as the most frequently updated one. However, the real time collection of data often results with issues with data quality – the problem that is further examined in Sect. 2.2. PSMSL, UHSLC, and SONEL databases are also actively maintained and receive regular updates. In comparison, GESLA is updated less often, with new releases occurring every 5 or 6 years (Menéndez and Woodworth, 2010; Woodworth et al., 2017; Haigh et al., 2022; GESLA-4, https://gesla787883612.wordpress.com/downloads/, last access: 15 June 2026) and the MISELA dataset covers a period of 2004–2019, and has not yet been updated.

In addition to the previously mentioned data portals and databases, sea level time series are also available through integrated repositories. Notable examples include the European Marine Observation and Data Network (EMODnet; https://emodnet.ec.europa.eu/en/physics, last access: 15 June 2026), which offers global coverage and the Copernicus Marine Environment Monitoring service in Situ TAC (CMEMS INS TAC; https://marineinsitu.eu/, last access: 15 June 2026) which primarily covers Europe. The tide gauge intercomparison tool, TGCAT (Tide Gauge CATalogs; https://www.sonel.org/tgcat, last access: 15 June 2026), serves as a valuable resource for the comparison of fundamental tide gauge information (e.g., coordinates, names) across diverse data repositories, especially given the increasing number of such repositories. Sea level data can be accessed, not only from global and regional databases, but also through the websites of national data providers. For instance, in the USA, sea level data measured at 1 min, and 6 min time step is freely available on the National Oceanic and Atmospheric Administration (NOAA) website (https://tidesandcurrents.noaa.gov/products.html, last access: 15 June 2026).

One of the main challenges with the databases mentioned earlier, as well as sea level measurements in general, is data quality and latency. Often, real time and near real time data, which are indispensable for navigation, port management, storm surge forecasting, and tsunami warning systems, are of unverified quality. In these situations, immediate data transmission and data sampling intervals of 1 min or less (IOC, 2006) are more important than rigorous quality control, which is often not feasible. However, real time monitoring can include simple checks such as detecting when a tide gauge stops transmitting data (UNESCO/IOC, 2020). Real time data are, e.g., available through the IOC SLSMF website and the National Tidal and Sea Level Facility (NTSLF) website (https://ntslf.org/data/uk-network-real-time, last access: 15 June 2026) for the UK Tide Gauge Network.

Higher quality data includes fast mode data, which is required on timescales of days to weeks, and is subject to mostly automatic quality control procedures, such as identifying and correcting obvious spikes. Such data is essential for the calibration of satellite altimeter data (Mitchum, 2000) and for forecasting ocean circulation and properties. These data are, e.g., available through the University of Hawaii Sea Level Center. On the other hand, delayed mode data is generally considered to be of the highest quality. It is primarily used for scientific research and subsequently for practical applications and is available at various time steps, through e.g., MISELA (for 1 min data, although containing only filtered sea level oscillations at periods shorter than 2 h), UHSLC, JASL, GESLA (all three primarily hourly data, with some higher resolution data in GESLA), PSMSL (monthly and annual data), and through national data providers.

Out of the listed datasets MISELA is the only one that contains 1 min sea level records that were thoroughly quality controlled (Zemunik et al., 2021). However, sea level records in MISELA contain only high-pass filtered series obtained after tidal signal was removed from the original series, and after the residual series were high-pass filtered with a 2 h Kaiser-Bessel window. Accordingly, quality control in MISELA is limited to high-frequency component. This methodological approach introduces a practical gap by restricting subsequent analysis strictly to short period sea level oscillations (T<2 h). Consequently, a comprehensive, sub-hourly quality-controlled dataset that preserves the total, unfiltered signal is essential. Such data is required to investigate both short- and long- period processes, as well as to quantify the contribution of the short-period oscillations to total sea level variations, particularly during extreme events.

While such high-resolution framework is ideally needed worldwide, managing a global scale was unfeasible to us given the resources required for the quality control of 1 min data, hence we decided to base SHELDA on the European sea level records. In addition to feasibility, choice of the European coast is supported by: (1) length and number of available records – longest and spatially most densely distributed time series of 1–15 min sea level data are available for the European and North America coasts; (2) extraordinary sea level variability, encompassing a wide range of tidal regimes as well as both low- and high- frequency oscillations is recorded along the European coast, allowing for research of multitude of processes of scientific and practical interest.

In this manuscript and related dataset, we present the delayed mode dataset of sea level series measured with a time step of 1–15 min along the European coasts and Greenland, Canary Islands, Israel, Lebanon and Türkiye. Our dataset originates entirely from the IOC SLSMF dataset of real-time sea level data. Prior to publishing the dataset, all data were subjected to strict automatic and visual quality control procedures. This was deemed necessary as real time data are prone to various quality issues, such as mean and date shifts, artificial spikes, gaps, and others (e.g., Šepić et al., 2015; Vilibić and Šepić, 2017; Woodworth, 2017; Zemunik et al., 2021). Ultimately, our objective was to establish a high-quality dataset that enables the precise identification and analysis of sea level variations occurring at periods from minutes to seasonal and longer, depending on the stability of long term series along the majority of the European coastline.

This paper is structured as follows. Section 2 describes the data source for the SHELDA dataset and outlines the quality control procedures implemented. Section 3 presents the dataset itself, detailing the spatial and temporal coverage of the quality-controlled time series, along with the technical specifications of the dataset. The paper subsequently includes a data availability statement and concludes with a discussion on the applications of the dataset and potential areas for improvement.

The dataset consists of 257 quality-controlled sea level time series in the NetCDF format, each representing data from an individual tide gauge station along the European coastline. Each file is named according to the station code specified on the IOC SLSMF website and contains four variables: (i) “time” – expressed in seconds since 1 January 2000 00:00:00 UTC; (ii) “sea_level_qc” – the quality-controlled sea level time series; (iii) “residual” – quality-controlled record minus the tidal signal; (iv) “qc_flags” – quality control flags, where 1 indicates high-quality data, 2 denotes missing data and 3 removed data. For records which are split into segments of various sampling frequencies all four variables are given separately for each of the segments.

The global attributes include metadata on the station name (IOC code and full station name), geographic location (country, latitude, and longitude), sampling interval and temporal coverage of the quality-controlled records. Additional attributes specify the mean value removed from the raw data, number of segments contained in each file, the original data source (i.e., the IOC SLSMF portal where data is hosted), the data provider (i.e., the data owner or the institution responsible for maintaining the station) and contact information of the dataset compiler. When applicable (e.g. in cases involving variable sampling frequencies in the raw data or corrections for vertical shifts), a disclaimer is included to explain the additional global attributes. A simplified example of the NetCDF file structure used in the SHELDA dataset is provided in Table 1.

SHELDA dataset contains nearly 1816 station years of sea level data of 1–15 min frequency, coming from 257 records across 22 different countries (Fig. 1). These records cover the period from 2 December 2007 to 1 January 2021, depending on station, with an overall average record length of 7.1 years. However, the duration of individual records varies significantly among stations, ranging from 1 month (amas, erdek, igne, ista, ment, sile and yalo in Türkiye) to as long as 13 years (smog, sthm and goto in Sweden).

The distribution of sea level monitoring stations (for which data is available in the IOC database) across European coastal regions shows significant geographical differences (Fig. 1). The United Kingdom, France, Spain, and Italy have high station coverage with records of various lengths. Despite their long coastline, Norway and Sweden show low station density in the IOC database. Similarly, along the coastlines of Ireland, Iceland, Portugal, Croatia, and the Black Sea, there is limited data availability within the IOC database. Notable gaps are evident across several European countries, with Finland, Estonia, Latvia, Lithuania, Poland, Slovenia, Montenegro, Albania, and Bulgaria completely absent from the IOC database, while Germany, although present in the IOC database, is excluded from the SHELDA dataset due to insufficient data quality (i.e. presence of numerous non-physical oscillations appearing throughout the time series). The lack of data from specific countries does not necessarily imply that sea level measurements are not conducted or are conducted only sparsely in these countries, but often that data (or parts of data) are not shared via the IOC SLSMF web portal. E.g., Pérez Gómez et al. (2022) in their study on the coastal sea level monitoring stations in the Mediterranean list 17 tide gauge stations measuring with 1 min time step along the Croatian coast of the Adriatic Sea – however, data from only three of these stations are available at the IOC SLSMF web portal.

Analysis of record durations shows that stations with longer records (over 10 years) are predominantly found in the Baltic region, France, Italy, and Spain. In contrast, those with shorter records (less than 4 years) are mainly located in the Eastern Mediterranean and the Black Sea. The longest records in the SHELDA dataset are from 3 tide gauges in Sweden. Stations with shorter time series may be newer installations, may have only operated for a short period, or may have been recently added to the IOC database. In some cases, records are shorter than the available raw data because numerous erroneous data points were removed during extensive quality control, particularly at the beginning and/or end of the records.

Besides spatial distribution, the following regional close-up views (Figs. 7–9) reveal detailed insights about the infrastructure supporting sea level monitoring. The figures show locations, technological preferences (i.e. type of sensor) and operational characteristics (i.e. sampling frequency) of stations in three regions: (a) Region 1: Central and Eastern Mediterranean Sea and Black Sea regions (98 stations; Fig. 7); (b) Region 2: Western Mediterranean, Western European coasts, and Canary Islands (84 stations; Fig. 8); (c) Region 3: Northern European coasts, including the United Kingdom, Ireland, Iceland, Greenland, Norway, Sweden, Denmark, the Netherlands, and Belgium (75 stations; Fig. 9).

The bar charts in Fig. 10 give an overview of the distribution of records across different countries, showing the cumulative years of data available for each country, the periods those records cover, the starting year of records, and specific technical information about the tide gauges featured in the dataset. As shown in Fig. 10a, Spain stands out as the top contributor, with 50 records and just over 400 years of accumulated data, closely followed by the United Kingdom. Italy and France contribute fewer records (30–37) but still exceed 270 years of accumulated data. When comparing the number of records to the total years of data collected for each country, significant differences are found. For instance, Türkiye and Greece have a relatively high number of records but shorter cumulative durations, indicating shorter records. Other countries have considerably less data, highlighting regional differences in data accessibility.

The temporal distribution of records shows variation in time series length (Fig. 10b); while some records span over a decade (altogether 68 records), there are also numerous records shorter than three years (altogether 55 records). Additionally, information on the starting year of records (Fig. 10c) provides important context for understanding the development and expansion of the IOC database over time. Peaks in 2008 and 2012 correspond to the addition of Italian stations (16 of 36) and UK stations (42 of 47) to the network, respectively.

From a technical point of view, a 1 min sampling frequency is the most common temporal resolution in the final processed dataset, accounting for 69.6 % of all records, as illustrated in Fig. 10d. This frequency aligns with modern oceanographic standards and international protocols. It is the dominant sampling frequency observed in Region 1 (Fig. 7) and Region 2 (Fig. 8). In contrast, in Region 3 (Fig. 9), instruments with sampling frequencies of 10 and 15 min are predominantly used. The final chart presents information about instrumentation choices and categorizes records based on the tide gauge type (Fig. 10e). A significant portion of the network is equipped with radar sensors, which account for 68.1 % of all sensors. This aligns with the IOC recommendations for upgrading tide gauge networks. The analysis of tide gauge types shows that radar sensors are the dominant choice in both Region 1 (Fig. 7) and Region 2 (Fig. 8), while bottom pressure recorders and float tide gauges are more commonly used in Region 3 (Fig. 9). There are also stations equipped with step gauges. A step gauge represents a specific type of tide gauge that measures water level using a vertical array of evenly spaced metal electrodes. It operates by electronically scanning the electrodes to identify the lowest dry one, which by definition marks the actual water level.

The distribution of sampling frequencies and sensor types in the SHELDA dataset reflects a combination of the hydrodynamic characteristics of the European sea basins, the primary objectives of national tide gauge networks, and their historical development. Different coastal regions require different temporal resolutions depending on the hydrodynamic processes that dominate them. In micro-tidal regions, such as the Mediterranean and Black Seas, high-frequency sampling (1 min or less) is necessary to capture short-period oscillations, including seiches and tsunamis, as well as meteotsunamis. In contrast, macro-tidal regions such as the North Sea and the English Channel have traditionally relied on 10 to 15 min sampling intervals, which are generally sufficient for capturing long-period oscillations, such as astronomical tides and storm surges.

These regional differences are closely linked to the development and modernization of national tide gauge systems. Monitoring networks in the Mediterranean and Black Sea regions (Regions 1 and 2) have undergone substantial upgrades in recent decades, largely driven by the implementation of tsunami warning systems since the mid-2000s. As a result, there has been a widespread transition toward radar sensors, particularly in countries such as Spain, France, Italy, Croatia and Türkiye (Pérez Gómez et al., 2022). By contrast, many Northern European networks (Region 3) continue to rely on established infrastructure, including float gauges in stilling wells and bottom pressure sensors. This persistence reflects their proven long-term reliability and resilience to harsh environmental conditions. Some national agencies plan to preserve existing sensors to ensure continuity of long-term data records, while simultaneously upgrading systems to modern standards (e.g., the United Kingdom's National Tide Gauge Network; https://www.oceanwise-global.com/article/oceanwise-to-maintain-and-upgrade-the-uks-national-tide-gauge- network/, last access: 26 June 2026).

Data described in this paper can be accessed through the Marine Data Archive of the Flanders Marine Institute (VLIZ) at 10.14284/764 (Balić and Šepić, 2025). This dataset is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).

The Sub-Hourly European Sea Level Dataset (SHELDA) provides a valuable resource for analysis of sea level variations occurring at periods from minutes to seasonal and longer, depending on the stability of long term series along European coastlines. The utilization of such data for research purposes requires the previous implementation of rigorous quality control procedures to ensure the prevention of misinterpretation of findings and dissemination of inaccurate conclusions. By performing quality control procedures on raw sea level time series from the IOC SLSMF database, we have created a high-quality research dataset containing nearly 1816 stations years of sea level data sampled with a 1–15 min time step and originating from 257 tide gauge stations across 22 European countries. The procedures used for quality control of the SHELDA dataset involved several steps that combined both automated and visual control methods. Automated methods were used so the same criteria could be easily and quickly applied across all data, and visual inspection had to be included because there were some erroneous data that the automated processes missed. Particularly challenging were spurious data points, such as flat lines and clustered spikes, and special issues like attenuated signals. Statistical analysis of the quality control process indicated that visual inspection remains essential despite automated procedures, accounting for 55.31 % of the total data removals. This underscores the limitation of our automated quality control algorithm and highlights the potential for developing techniques that will automatically detect the aforementioned problems. The development of those advanced algorithms will significantly improve and accelerate quality control, making visual inspection less time consuming compared to our previous experience.

A comparison of SHELDA with similar existing datasets, GESLA (Haigh et al., 2022) and MISELA (Zemunik et al., 2021), shows the differences between them. While GESLA mainly consists of global hourly sea level measurements with different levels of quality control, SHELDA focuses on European sea level records sampled with a 1–15 min time step, implementing rigorous quality control procedures. Compared to MISELA, which provides 1 min high-pass (2 h cut-off period) quality-controlled records, SHELDA is more appropriate for the analysis of sea level processes across a wider range of timescales because it preserves the full spectral content of measurements. Access to high-quality sea level data of a 1–15 min time step is important for improving our understanding of coastal flooding hazards, which often result from complex interactions between different sea level components across multiple timescales (Woodworth et al., 2019; Idier et al., 2019). The SHELDA dataset is particularly valuable for investigating contribution of high-frequency phenomena to flooding.

Several limitations of the SHELDA dataset must be acknowledged. First, the spatial distribution of tide gauges remains uneven across the European coasts. Expanding geographical coverage could involve adding more stations in underrepresented regions, which might require incorporating additional data sources beyond the IOC SLSMF database and collaborating with regional monitoring programs to improve data sharing. Second, the temporal coverage of the dataset is currently limited to the end of 2020, primarily due to the extensive time and effort required for performing rigorous quality control, particularly the visual inspection step. We recognize that the inclusion of more recent data would significantly enhance the value of SHELDA so we plan to extend the dataset in the future as resources allow. There is also significant potential for expansion, as at least 70 new stations have been added to the IOC database since January 2021 (e.g., 50 Swedish and 21 Norwegian sea level stations, added to the system in October 2021 and June 2023, respectively). Third, the temporal coverage varies significantly between stations, with 55 of them having records that span less than three years (up to 2021). This is particularly evident for the Croatian Adriatic coast and the eastern Mediterranean and Black Sea regions. While the availability of short time series, especially when they are new records, is welcome, short series resulting from discontinued tide gauge operations are concerning. Therefore, in addition to recommending the targeted addition of new tide gauges to the European sea level monitoring network and advocating for open data and free data access where data exists but is not shared due to local regulations (e.g., Pérez Gómez et al., 2022 for the Mediterranean), it is also necessary to focus on the maintenance of existing tide gauges to ensure that they continue to operate effectively in the future.

To ensure that the dataset remains useful for both operational and research purposes, it will need to be updated regularly. Addressing the current temporal limitation will depend largely on improving the efficiency of the quality control workflow. In this context, the development of more advanced automated quality control methods would be particularly helpful. A promising step in this direction is the VLIZ API (Application Programming Interface) web service for the SLSMF, introduced in spring 2025, which provides access to automatically quality-controlled sea level data (https://api.ioc-sealevelmonitoring.org/v2/doc, last access: 15 June 2026). Using this service could significantly simplify and accelerate our data processing workflow, as the automated quality control methodology is compatible with our existing procedures. However, it is important to note that the API does not include visual inspection, so, if the highest data quality is required, an additional visual quality control step may still be necessary.

Extreme sea levels (Vousdoukas et al., 2017) and coastal flooding (IPCC, 2014; Hinkel et al., 2014) are growing threat to European coastlines, largely due to rising mean sea levels driven by climate change (Oppenheimer et al., 2019; Hamlington et al., 2024). In recent decades, the frequency of extreme sea level events has increased at many locations (Menéndez and Woodworth, 2010; Haigh et al., 2010). To predict these hazards, researchers use global models of tide, storm surges, and wave setup to obtain projections of episodic coastal flooding over the coming century (Kirezci et al., 2020). These models are validated against tide gauge data, highlighting the critical need for reliable, high-quality datasets like SHELDA. By sharing this quality-controlled dataset openly with researchers, we hope to make it easier to analyse the complex sea level phenomena behind coastal flooding, particularly in regions where high-frequency processes may dominate or significantly contribute to overall extremes. This improved ability to study processes across different timescales will be crucial for creating effective adaptation strategies and helping coastal areas become more resilient.