Jadranka Šepić
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
Jadranka Šepić has a total of 208 co-authors. Here are the top co-authors by number of shared publications:
- Ivica Vilibić
Ruđer Bošković Institute (Zagreb, Croatia)
25 shared publications Ivica Vilibić
21 shared publications- Krešimir Ruić
Faculty of Science, University of Split, Split, Croatia
20 shared publications Marin Vojković
Faculty of Science, University of Split, 21000 Split, Croatia
18 shared publications- Iva Međugorac
University of Nova Gorica, Vipavska 13, Rozna Dolina, SI-5000 Nova Gorica, Slovenia
13 shared publications Hrvoje Mihanović
12 shared publications
Latest Publications
- Open Access
- 27 October 2026
- Published by Springer Nature in Pure and Applied Geophysics
- p. 1-43
Spatio-Temporal Analysis of Hydrological Extremes, Waves and Storm Surges During the 2013–14 Winter Storms, UK
The 2013–14 UK winter storm series was characterised by an exceptional frequency and persistence of storm events, with impacts arising from cumulative hydro-meteorological forcing rather than the intensity of individual storms. Sustained atmospheric forcing resulted in ten fatalities, extensive flooding, coastal erosion, and damage to critical infrastructure. This study investigates how the evolution of atmospheric forcing during the 2013–14 winter influenced coastal ocean variability and impacts across the UK, using a consistent observational framework developed from a decade of additional research. A storm and flood chronology and impacts are delineated in a comprehensive review, demonstrating how repeated storm clusters prevented recovery of coastal systems and amplified impacts. An official direct damage estimate of £1300 M likely substantially underestimated the cost of the storms by neglecting indirect impacts from disruptions. Analysis disaggregates these costs by sector, highlighting damage to residential properties, businesses, highways, and rail infrastructure. The Dawlish railway line collapse and the sustained flooding of the Somerset Levels are presented. Post-storm recovery is noted to be variable, reflecting differences in local coastal processes, infrastructure resilience, and repeated storm events. A synoptic dataset comprising waves, tide gauges, wind, and precipitation is analysed with data sourced from 44 tide gauges, 52 wave buoys, and 15 weather station records. Results showed tide gauges and wave buoys exceeded recurrence thresholds of storm surge and wave height at a greater frequency than other winter seasons, demonstrating the disproportional occurrence of extreme waves. The maximum storm surge was recorded at Lowestoft during Storm Xaver at ∼2.2 m, which contributed to the extensive coastal damage on the east coast. Persistent rainfall compounded impacts, with South England and Wales receiving 200–300% of average winter precipitation. This research demonstrates how the sequencing of storms and the interaction of coastal, meteorological, and hydrological extremes generated cumulative impacts that overwhelmed natural and engineered defences. While direct attribution of the 2013–14 winter impacts to anthropogenic climate change remains indeterminate, UK Climate Projections (UKCP18) indicate wetter winters and increased flood risk, emphasising the need for integrated risk assessment, resilient infrastructure, and adaptive strategies for managing prolonged multi-hazard events.
- 27 October 2026
- Published by Springer Nature in Pure and Applied Geophysics
- p. 1-43
Open AccessSpatio-Temporal Analysis of Hydrological Extremes, Waves and Storm Surges During the 2013–14 Winter Storms, UK
The 2013–14 UK winter storm series was characterised by an exceptional frequency and persistence of storm events, with impacts arising from cumulative hydro-meteorological forcing rather than the intensity of individual storms. Sustained atmospheric forcing resulted in ten fatalities, extensive flooding, coastal erosion, and damage to critical infrastructure. This study investigates how the evolution of atmospheric forcing during the 2013–14 winter influenced coastal ocean variability and impacts across the UK, using a consistent observational framework developed from a decade of additional research. A storm and flood chronology and impacts are delineated in a comprehensive review, demonstrating how repeated storm clusters prevented recovery of coastal systems and amplified impacts. An official direct damage estimate of £1300 M likely substantially underestimated the cost of the storms by neglecting indirect impacts from disruptions. Analysis disaggregates these costs by sector, highlighting damage to residential properties, businesses, highways, and rail infrastructure. The Dawlish railway line collapse and the sustained flooding of the Somerset Levels are presented. Post-storm recovery is noted to be variable, reflecting differences in local coastal processes, infrastructure resilience, and repeated storm events. A synoptic dataset comprising waves, tide gauges, wind, and precipitation is analysed with data sourced from 44 tide gauges, 52 wave buoys, and 15 weather station records. Results showed tide gauges and wave buoys exceeded recurrence thresholds of storm surge and wave height at a greater frequency than other winter seasons, demonstrating the disproportional occurrence of extreme waves. The maximum storm surge was recorded at Lowestoft during Storm Xaver at ∼2.2 m, which contributed to the extensive coastal damage on the east coast. Persistent rainfall compounded impacts, with South England and Wales receiving 200–300% of average winter precipitation. This research demonstrates how the sequencing of storms and the interaction of coastal, meteorological, and hydrological extremes generated cumulative impacts that overwhelmed natural and engineered defences. While direct attribution of the 2013–14 winter impacts to anthropogenic climate change remains indeterminate, UK Climate Projections (UKCP18) indicate wetter winters and increased flood risk, emphasising the need for integrated risk assessment, resilient infrastructure, and adaptive strategies for managing prolonged multi-hazard events.
- Open Access
- 12 September 2026
- Published by Copernicus GmbH
- Vol. 2026, 1-31
Reply on RC2
The eastern Adriatic coast is a known hotspot of strong meteorologically induced high-frequency sea-level oscillations, occurring at periods shorter than 1 hour and reaching wave heights of several metres. When highest, these oscillations are termed meteotsunamis. In this study, we test deep-learning methods for predicting maximum daily amplitudes of high-frequency (T < 1 hour) sea-level oscillations at two Adriatic locations, Bakar and Ploče, using convolutional neural networks driven by past sea-level observations and atmospheric predictors from the ERA5 and CERRA reanalyses. We evaluate two deep-learning architectures designed to test different approaches to representing sea-level and atmospheric forcing. The first architecture, HFNet, is based on the HIDRA family of models, a general low-frequency sea-level forecasting framework that has been extensively evaluated in the Adriatic and shown to provide a credible baseline for sea-level prediction. The second architecture, HFNetJE, extends this approach through joint encoding of atmospheric predictors and a more extensive processing of past sea-level information, with the aim of improving the representation of processes associated with high-frequency sea-level oscillations. Analysis of more than 20 years of data shows that high-frequency sea-level extremes are larger in Bakar (> 60 cm) than in Ploče (< 35 cm), occur ~6 times per year, and are most common during the warm season. Both architectures reproduce the observed variability, with higher skill for typical than for extreme events. HFNetJE performs best overall and under typical amplitude conditions, whereas HFNet more effectively captures extreme events, although these remain systematically underestimated in both architectures. Model performance is higher at Ploče, likely because of its smaller sea-level range and simpler response to atmospheric forcing. Models forced with ERA5 consistently outperform those using the higher-resolution CERRA in predicting extremes, suggesting limited added value from increased spatial resolution. Ablation experiments indicate that several predictors are redundant for average forecasting performance, whereas extreme-event prediction generally benefits from the full predictor set. Overall, the results demonstrate the potential of deep learning for prediction of high-frequency sea-level oscillations in the Adriatic, but also highlight persistent limitations in forecasting rare high-amplitude events.
- 12 September 2026
- Published by Copernicus GmbH
- Vol. 2026, 1-31
Open AccessReply on RC2
The eastern Adriatic coast is a known hotspot of strong meteorologically induced high-frequency sea-level oscillations, occurring at periods shorter than 1 hour and reaching wave heights of several metres. When highest, these oscillations are termed meteotsunamis. In this study, we test deep-learning methods for predicting maximum daily amplitudes of high-frequency (T < 1 hour) sea-level oscillations at two Adriatic locations, Bakar and Ploče, using convolutional neural networks driven by past sea-level observations and atmospheric predictors from the ERA5 and CERRA reanalyses. We evaluate two deep-learning architectures designed to test different approaches to representing sea-level and atmospheric forcing. The first architecture, HFNet, is based on the HIDRA family of models, a general low-frequency sea-level forecasting framework that has been extensively evaluated in the Adriatic and shown to provide a credible baseline for sea-level prediction. The second architecture, HFNetJE, extends this approach through joint encoding of atmospheric predictors and a more extensive processing of past sea-level information, with the aim of improving the representation of processes associated with high-frequency sea-level oscillations. Analysis of more than 20 years of data shows that high-frequency sea-level extremes are larger in Bakar (> 60 cm) than in Ploče (< 35 cm), occur ~6 times per year, and are most common during the warm season. Both architectures reproduce the observed variability, with higher skill for typical than for extreme events. HFNetJE performs best overall and under typical amplitude conditions, whereas HFNet more effectively captures extreme events, although these remain systematically underestimated in both architectures. Model performance is higher at Ploče, likely because of its smaller sea-level range and simpler response to atmospheric forcing. Models forced with ERA5 consistently outperform those using the higher-resolution CERRA in predicting extremes, suggesting limited added value from increased spatial resolution. Ablation experiments indicate that several predictors are redundant for average forecasting performance, whereas extreme-event prediction generally benefits from the full predictor set. Overall, the results demonstrate the potential of deep learning for prediction of high-frequency sea-level oscillations in the Adriatic, but also highlight persistent limitations in forecasting rare high-amplitude events.
- Open Access
- 12 September 2026
- Published by Copernicus GmbH
- Vol. 2026, 1-31
Reply on RC1
The eastern Adriatic coast is a known hotspot of strong meteorologically induced high-frequency sea-level oscillations, occurring at periods shorter than 1 hour and reaching wave heights of several metres. When highest, these oscillations are termed meteotsunamis. In this study, we test deep-learning methods for predicting maximum daily amplitudes of high-frequency (T < 1 hour) sea-level oscillations at two Adriatic locations, Bakar and Ploče, using convolutional neural networks driven by past sea-level observations and atmospheric predictors from the ERA5 and CERRA reanalyses. We evaluate two deep-learning architectures designed to test different approaches to representing sea-level and atmospheric forcing. The first architecture, HFNet, is based on the HIDRA family of models, a general low-frequency sea-level forecasting framework that has been extensively evaluated in the Adriatic and shown to provide a credible baseline for sea-level prediction. The second architecture, HFNetJE, extends this approach through joint encoding of atmospheric predictors and a more extensive processing of past sea-level information, with the aim of improving the representation of processes associated with high-frequency sea-level oscillations. Analysis of more than 20 years of data shows that high-frequency sea-level extremes are larger in Bakar (> 60 cm) than in Ploče (< 35 cm), occur ~6 times per year, and are most common during the warm season. Both architectures reproduce the observed variability, with higher skill for typical than for extreme events. HFNetJE performs best overall and under typical amplitude conditions, whereas HFNet more effectively captures extreme events, although these remain systematically underestimated in both architectures. Model performance is higher at Ploče, likely because of its smaller sea-level range and simpler response to atmospheric forcing. Models forced with ERA5 consistently outperform those using the higher-resolution CERRA in predicting extremes, suggesting limited added value from increased spatial resolution. Ablation experiments indicate that several predictors are redundant for average forecasting performance, whereas extreme-event prediction generally benefits from the full predictor set. Overall, the results demonstrate the potential of deep learning for prediction of high-frequency sea-level oscillations in the Adriatic, but also highlight persistent limitations in forecasting rare high-amplitude events.
- 12 September 2026
- Published by Copernicus GmbH
- Vol. 2026, 1-31
Open AccessReply on RC1
The eastern Adriatic coast is a known hotspot of strong meteorologically induced high-frequency sea-level oscillations, occurring at periods shorter than 1 hour and reaching wave heights of several metres. When highest, these oscillations are termed meteotsunamis. In this study, we test deep-learning methods for predicting maximum daily amplitudes of high-frequency (T < 1 hour) sea-level oscillations at two Adriatic locations, Bakar and Ploče, using convolutional neural networks driven by past sea-level observations and atmospheric predictors from the ERA5 and CERRA reanalyses. We evaluate two deep-learning architectures designed to test different approaches to representing sea-level and atmospheric forcing. The first architecture, HFNet, is based on the HIDRA family of models, a general low-frequency sea-level forecasting framework that has been extensively evaluated in the Adriatic and shown to provide a credible baseline for sea-level prediction. The second architecture, HFNetJE, extends this approach through joint encoding of atmospheric predictors and a more extensive processing of past sea-level information, with the aim of improving the representation of processes associated with high-frequency sea-level oscillations. Analysis of more than 20 years of data shows that high-frequency sea-level extremes are larger in Bakar (> 60 cm) than in Ploče (< 35 cm), occur ~6 times per year, and are most common during the warm season. Both architectures reproduce the observed variability, with higher skill for typical than for extreme events. HFNetJE performs best overall and under typical amplitude conditions, whereas HFNet more effectively captures extreme events, although these remain systematically underestimated in both architectures. Model performance is higher at Ploče, likely because of its smaller sea-level range and simpler response to atmospheric forcing. Models forced with ERA5 consistently outperform those using the higher-resolution CERRA in predicting extremes, suggesting limited added value from increased spatial resolution. Ablation experiments indicate that several predictors are redundant for average forecasting performance, whereas extreme-event prediction generally benefits from the full predictor set. Overall, the results demonstrate the potential of deep learning for prediction of high-frequency sea-level oscillations in the Adriatic, but also highlight persistent limitations in forecasting rare high-amplitude events.
- Open Access
- 26 August 2026
- Published by Springer Nature in Natural Hazards
- Vol. 122 (18) , 616
Investigation of North Sea high-frequency sea-level extremes using long-term sea-level measurements
Extreme sea-level events occur across a range of temporal and spatial scales, including high-frequency oscillations (periods T < 2 h). However, most research uses hourly or daily data and neglects higher-frequency processes such as seiches and meteotsunamis. This study investigates the characteristics of high-frequency (HF) sea-level extremes along the North Sea coast. Long-term (1993–2025) measurements were analysed from 29 tide gauges. After quality control, the astronomical tide was removed to isolate residuals, which were decomposed into low-frequency (T > 2 h) and high-frequency (T < 2 h) components. Next, extremes were extracted from the HF signal alone (HF extremes) and from the residual (residual extremes). Clustering techniques (K-medoids with Dynamic-Time-Warping and Euclidean distances) were applied to HF extremes to classify event types and identify regional patterns. HF extremes were grouped into six event types, and selected stations yielded five spatial clusters. Each event type was characterised by period, intensity, and its relationship to residual extremes. HF extremes were generally low compared with tidal ranges, and compound events (i.e., events in which an HF extreme coincided with a residual extreme) were infrequent (12.2% of HF extremes on average). Nevertheless, during compound events, alignment of HF and residual peaks substantially amplified sea levels. Spatial patterns highlighted the most intense events along the Dutch coast (IJmuiden Buitenhaven, Scheveningen, Brouwershavensche Gat, and Terschelling). These results were influenced by differences in local environmental conditions and sampling. Therefore, this study advances understanding of HF sea-level variability and compound events along the North Sea coast, supporting improved coastal hazard assessments.
- 26 August 2026
- Published by Springer Nature in Natural Hazards
- Vol. 122 (18) , 616
Open AccessInvestigation of North Sea high-frequency sea-level extremes using long-term sea-level measurements
Extreme sea-level events occur across a range of temporal and spatial scales, including high-frequency oscillations (periods T < 2 h). However, most research uses hourly or daily data and neglects higher-frequency processes such as seiches and meteotsunamis. This study investigates the characteristics of high-frequency (HF) sea-level extremes along the North Sea coast. Long-term (1993–2025) measurements were analysed from 29 tide gauges. After quality control, the astronomical tide was removed to isolate residuals, which were decomposed into low-frequency (T > 2 h) and high-frequency (T < 2 h) components. Next, extremes were extracted from the HF signal alone (HF extremes) and from the residual (residual extremes). Clustering techniques (K-medoids with Dynamic-Time-Warping and Euclidean distances) were applied to HF extremes to classify event types and identify regional patterns. HF extremes were grouped into six event types, and selected stations yielded five spatial clusters. Each event type was characterised by period, intensity, and its relationship to residual extremes. HF extremes were generally low compared with tidal ranges, and compound events (i.e., events in which an HF extreme coincided with a residual extreme) were infrequent (12.2% of HF extremes on average). Nevertheless, during compound events, alignment of HF and residual peaks substantially amplified sea levels. Spatial patterns highlighted the most intense events along the Dutch coast (IJmuiden Buitenhaven, Scheveningen, Brouwershavensche Gat, and Terschelling). These results were influenced by differences in local environmental conditions and sampling. Therefore, this study advances understanding of HF sea-level variability and compound events along the North Sea coast, supporting improved coastal hazard assessments.
- Open Access
- 31 July 2026
- Published by Copernicus GmbH
- Vol. 2026, 1-31
Comment on egusphere-2026-3411
The eastern Adriatic coast is a known hotspot of strong meteorologically induced high-frequency sea-level oscillations, occurring at periods shorter than 1 hour and reaching wave heights of several metres. When highest, these oscillations are termed meteotsunamis. In this study, we test deep-learning methods for predicting maximum daily amplitudes of high-frequency (T < 1 hour) sea-level oscillations at two Adriatic locations, Bakar and Ploče, using convolutional neural networks driven by past sea-level observations and atmospheric predictors from the ERA5 and CERRA reanalyses. We evaluate two deep-learning architectures designed to test different approaches to representing sea-level and atmospheric forcing. The first architecture, HFNet, is based on the HIDRA family of models, a general low-frequency sea-level forecasting framework that has been extensively evaluated in the Adriatic and shown to provide a credible baseline for sea-level prediction. The second architecture, HFNetJE, extends this approach through joint encoding of atmospheric predictors and a more extensive processing of past sea-level information, with the aim of improving the representation of processes associated with high-frequency sea-level oscillations. Analysis of more than 20 years of data shows that high-frequency sea-level extremes are larger in Bakar (> 60 cm) than in Ploče (< 35 cm), occur ~6 times per year, and are most common during the warm season. Both architectures reproduce the observed variability, with higher skill for typical than for extreme events. HFNetJE performs best overall and under typical amplitude conditions, whereas HFNet more effectively captures extreme events, although these remain systematically underestimated in both architectures. Model performance is higher at Ploče, likely because of its smaller sea-level range and simpler response to atmospheric forcing. Models forced with ERA5 consistently outperform those using the higher-resolution CERRA in predicting extremes, suggesting limited added value from increased spatial resolution. Ablation experiments indicate that several predictors are redundant for average forecasting performance, whereas extreme-event prediction generally benefits from the full predictor set. Overall, the results demonstrate the potential of deep learning for prediction of high-frequency sea-level oscillations in the Adriatic, but also highlight persistent limitations in forecasting rare high-amplitude events.
- 31 July 2026
- Published by Copernicus GmbH
- Vol. 2026, 1-31
Open AccessComment on egusphere-2026-3411
The eastern Adriatic coast is a known hotspot of strong meteorologically induced high-frequency sea-level oscillations, occurring at periods shorter than 1 hour and reaching wave heights of several metres. When highest, these oscillations are termed meteotsunamis. In this study, we test deep-learning methods for predicting maximum daily amplitudes of high-frequency (T < 1 hour) sea-level oscillations at two Adriatic locations, Bakar and Ploče, using convolutional neural networks driven by past sea-level observations and atmospheric predictors from the ERA5 and CERRA reanalyses. We evaluate two deep-learning architectures designed to test different approaches to representing sea-level and atmospheric forcing. The first architecture, HFNet, is based on the HIDRA family of models, a general low-frequency sea-level forecasting framework that has been extensively evaluated in the Adriatic and shown to provide a credible baseline for sea-level prediction. The second architecture, HFNetJE, extends this approach through joint encoding of atmospheric predictors and a more extensive processing of past sea-level information, with the aim of improving the representation of processes associated with high-frequency sea-level oscillations. Analysis of more than 20 years of data shows that high-frequency sea-level extremes are larger in Bakar (> 60 cm) than in Ploče (< 35 cm), occur ~6 times per year, and are most common during the warm season. Both architectures reproduce the observed variability, with higher skill for typical than for extreme events. HFNetJE performs best overall and under typical amplitude conditions, whereas HFNet more effectively captures extreme events, although these remain systematically underestimated in both architectures. Model performance is higher at Ploče, likely because of its smaller sea-level range and simpler response to atmospheric forcing. Models forced with ERA5 consistently outperform those using the higher-resolution CERRA in predicting extremes, suggesting limited added value from increased spatial resolution. Ablation experiments indicate that several predictors are redundant for average forecasting performance, whereas extreme-event prediction generally benefits from the full predictor set. Overall, the results demonstrate the potential of deep learning for prediction of high-frequency sea-level oscillations in the Adriatic, but also highlight persistent limitations in forecasting rare high-amplitude events.
- Open Access
- 13 July 2026
- Published by Copernicus GmbH
- p. 1-31
Comment on egusphere-2026-3411
The eastern Adriatic coast is a known hotspot of strong meteorologically induced high-frequency sea-level oscillations, occurring at periods shorter than 1 hour and reaching wave heights of several metres. When highest, these oscillations are termed meteotsunamis. In this study, we test deep-learning methods for predicting maximum daily amplitudes of high-frequency (T < 1 hour) sea-level oscillations at two Adriatic locations, Bakar and Ploče, using convolutional neural networks driven by past sea-level observations and atmospheric predictors from the ERA5 and CERRA reanalyses. We evaluate two deep-learning architectures designed to test different approaches to representing sea-level and atmospheric forcing. The first architecture, HFNet, is based on the HIDRA family of models, a general low-frequency sea-level forecasting framework that has been extensively evaluated in the Adriatic and shown to provide a credible baseline for sea-level prediction. The second architecture, HFNetJE, extends this approach through joint encoding of atmospheric predictors and a more extensive processing of past sea-level information, with the aim of improving the representation of processes associated with high-frequency sea-level oscillations. Analysis of more than 20 years of data shows that high-frequency sea-level extremes are larger in Bakar (> 60 cm) than in Ploče (< 35 cm), occur ~6 times per year, and are most common during the warm season. Both architectures reproduce the observed variability, with higher skill for typical than for extreme events. HFNetJE performs best overall and under typical amplitude conditions, whereas HFNet more effectively captures extreme events, although these remain systematically underestimated in both architectures. Model performance is higher at Ploče, likely because of its smaller sea-level range and simpler response to atmospheric forcing. Models forced with ERA5 consistently outperform those using the higher-resolution CERRA in predicting extremes, suggesting limited added value from increased spatial resolution. Ablation experiments indicate that several predictors are redundant for average forecasting performance, whereas extreme-event prediction generally benefits from the full predictor set. Overall, the results demonstrate the potential of deep learning for prediction of high-frequency sea-level oscillations in the Adriatic, but also highlight persistent limitations in forecasting rare high-amplitude events.
- 13 July 2026
- Published by Copernicus GmbH
- p. 1-31
Open AccessComment on egusphere-2026-3411
The eastern Adriatic coast is a known hotspot of strong meteorologically induced high-frequency sea-level oscillations, occurring at periods shorter than 1 hour and reaching wave heights of several metres. When highest, these oscillations are termed meteotsunamis. In this study, we test deep-learning methods for predicting maximum daily amplitudes of high-frequency (T < 1 hour) sea-level oscillations at two Adriatic locations, Bakar and Ploče, using convolutional neural networks driven by past sea-level observations and atmospheric predictors from the ERA5 and CERRA reanalyses. We evaluate two deep-learning architectures designed to test different approaches to representing sea-level and atmospheric forcing. The first architecture, HFNet, is based on the HIDRA family of models, a general low-frequency sea-level forecasting framework that has been extensively evaluated in the Adriatic and shown to provide a credible baseline for sea-level prediction. The second architecture, HFNetJE, extends this approach through joint encoding of atmospheric predictors and a more extensive processing of past sea-level information, with the aim of improving the representation of processes associated with high-frequency sea-level oscillations. Analysis of more than 20 years of data shows that high-frequency sea-level extremes are larger in Bakar (> 60 cm) than in Ploče (< 35 cm), occur ~6 times per year, and are most common during the warm season. Both architectures reproduce the observed variability, with higher skill for typical than for extreme events. HFNetJE performs best overall and under typical amplitude conditions, whereas HFNet more effectively captures extreme events, although these remain systematically underestimated in both architectures. Model performance is higher at Ploče, likely because of its smaller sea-level range and simpler response to atmospheric forcing. Models forced with ERA5 consistently outperform those using the higher-resolution CERRA in predicting extremes, suggesting limited added value from increased spatial resolution. Ablation experiments indicate that several predictors are redundant for average forecasting performance, whereas extreme-event prediction generally benefits from the full predictor set. Overall, the results demonstrate the potential of deep learning for prediction of high-frequency sea-level oscillations in the Adriatic, but also highlight persistent limitations in forecasting rare high-amplitude events.
- Open Access
- 25 June 2026
- Published by Copernicus GmbH in EGUsphere
Deep-learning prediction of high-frequency sea-level oscillations in the Adriatic Sea
The eastern Adriatic coast is a known hotspot of strong meteorologically induced high-frequency sea-level oscillations, occurring at periods shorter than 1 hour and reaching wave heights of several metres. When highest, these oscillations are termed meteotsunamis. In this study, we test deep-learning methods for predicting maximum daily amplitudes of high-frequency (T < 1 hour) sea-level oscillations at two Adriatic locations, Bakar and Ploče, using convolutional neural networks driven by past sea-level observations and atmospheric predictors from the ERA5 and CERRA reanalyses. We evaluate two deep-learning architectures designed to test different approaches to representing sea-level and atmospheric forcing. The first architecture, HFNet, is based on the HIDRA family of models, a general low-frequency sea-level forecasting framework that has been extensively evaluated in the Adriatic and shown to provide a credible baseline for sea-level prediction. The second architecture, HFNetJE, extends this approach through joint encoding of atmospheric predictors and a more extensive processing of past sea-level information, with the aim of improving the representation of processes associated with high-frequency sea-level oscillations. Analysis of more than 20 years of data shows that high-frequency sea-level extremes are larger in Bakar (> 60 cm) than in Ploče (< 35 cm), occur ~6 times per year, and are most common during the warm season. Both architectures reproduce the observed variability, with higher skill for typical than for extreme events. HFNetJE performs best overall and under typical amplitude conditions, whereas HFNet more effectively captures extreme events, although these remain systematically underestimated in both architectures. Model performance is higher at Ploče, likely because of its smaller sea-level range and simpler response to atmospheric forcing. Models forced with ERA5 consistently outperform those using the higher-resolution CERRA in predicting extremes, suggesting limited added value from increased spatial resolution. Ablation experiments indicate that several predictors are redundant for average forecasting performance, whereas extreme-event prediction generally benefits from the full predictor set. Overall, the results demonstrate the potential of deep learning for prediction of high-frequency sea-level oscillations in the Adriatic, but also highlight persistent limitations in forecasting rare high-amplitude events.
- 25 June 2026
- Published by Copernicus GmbH in EGUsphere
Open AccessDeep-learning prediction of high-frequency sea-level oscillations in the Adriatic Sea
The eastern Adriatic coast is a known hotspot of strong meteorologically induced high-frequency sea-level oscillations, occurring at periods shorter than 1 hour and reaching wave heights of several metres. When highest, these oscillations are termed meteotsunamis. In this study, we test deep-learning methods for predicting maximum daily amplitudes of high-frequency (T < 1 hour) sea-level oscillations at two Adriatic locations, Bakar and Ploče, using convolutional neural networks driven by past sea-level observations and atmospheric predictors from the ERA5 and CERRA reanalyses. We evaluate two deep-learning architectures designed to test different approaches to representing sea-level and atmospheric forcing. The first architecture, HFNet, is based on the HIDRA family of models, a general low-frequency sea-level forecasting framework that has been extensively evaluated in the Adriatic and shown to provide a credible baseline for sea-level prediction. The second architecture, HFNetJE, extends this approach through joint encoding of atmospheric predictors and a more extensive processing of past sea-level information, with the aim of improving the representation of processes associated with high-frequency sea-level oscillations. Analysis of more than 20 years of data shows that high-frequency sea-level extremes are larger in Bakar (> 60 cm) than in Ploče (< 35 cm), occur ~6 times per year, and are most common during the warm season. Both architectures reproduce the observed variability, with higher skill for typical than for extreme events. HFNetJE performs best overall and under typical amplitude conditions, whereas HFNet more effectively captures extreme events, although these remain systematically underestimated in both architectures. Model performance is higher at Ploče, likely because of its smaller sea-level range and simpler response to atmospheric forcing. Models forced with ERA5 consistently outperform those using the higher-resolution CERRA in predicting extremes, suggesting limited added value from increased spatial resolution. Ablation experiments indicate that several predictors are redundant for average forecasting performance, whereas extreme-event prediction generally benefits from the full predictor set. Overall, the results demonstrate the potential of deep learning for prediction of high-frequency sea-level oscillations in the Adriatic, but also highlight persistent limitations in forecasting rare high-amplitude events.
- Open Access
- 25 June 2026
- Published by Copernicus GmbH
Supplementary material to "Deep-learning prediction of high-frequency sea-level oscillations in the Adriatic Sea"
- 25 June 2026
- Published by Copernicus GmbH
Open AccessSupplementary material to "Deep-learning prediction of high-frequency sea-level oscillations in the Adriatic Sea"
- Open Access
- 22 June 2026
- Published by Copernicus GmbH
Vertical Structure of Coastal Wind Regimes: Four Years of Continuous SODAR Observations at Split Airport
Split International Airport (LDSP), situated on the eastern Adriatic coast, presents a complex meteorological environment for aviation operations. The airport's location at the base of the Dinaric Alps creates a unique intersection between synoptic-scale flows and sub-mesoscale coastal circulations.Following a collaboration between the University of Split, Split Airport, and Croatia Control (air navigation service provider) continuous SODAR (SOnic Detection And Ranging) measurements have been operational since June 2022 at the airport location. The primary objective of this university project was the collection of research data, but since it was installed at the airport, it’s real-time measurements can also provide very useful information for aviation forecasters.The data is collected using a Scintec MFAS (Multi-Frequency Flat Array SODAR). While the instrument is capable of a vertical range up to 1000 m, it is optimized for high-resolution monitoring within the lowest 500 m of the troposphere—the most critical zone for aircraft approach and departure. Data is processed at 20-minute intervals, providing a detailed vertical profile of wind direction and speed.The four-year dataset captures the seasonal and diurnal variability of the four dominant local wind regimes:Bora (Bura): A gusty north-easterly wind, often associated with severe mechanical turbulence due to the nearby mountains.Sirocco (Jugo): A moist, south-easterly flow, sometimes with wind speeds above 25m/s in the 500m surface layer.Sea-Breeze (Maestral): A predictable but significant south-westerly coastal circulation that dictates runway changes from standard instrumental approach to a more complex visual approach from north-east, due to mountain proximity on that side of the airport.North-westerly gap flow : Katabatic flow from the mountain gap north-west of the airport, but also a tramontana, typical post-frontal flow (synoptic + orography).Moreover, in aviation meteorology information from SODAR instrument can be very useful for detecting vertical wind shear.This poster will present some typical examples of wind profiles at Split airport, wind roses at different altitudes (e.g., 50m, 150m, 300m), and examples of significant wind shear episodes.These findings emphasize the necessity of ground-based remote sensing for enhancing safety and operational efficiency at topographically challenged coastal airports.
- 22 June 2026
- Published by Copernicus GmbH
Open AccessVertical Structure of Coastal Wind Regimes: Four Years of Continuous SODAR Observations at Split Airport
Split International Airport (LDSP), situated on the eastern Adriatic coast, presents a complex meteorological environment for aviation operations. The airport's location at the base of the Dinaric Alps creates a unique intersection between synoptic-scale flows and sub-mesoscale coastal circulations.Following a collaboration between the University of Split, Split Airport, and Croatia Control (air navigation service provider) continuous SODAR (SOnic Detection And Ranging) measurements have been operational since June 2022 at the airport location. The primary objective of this university project was the collection of research data, but since it was installed at the airport, it’s real-time measurements can also provide very useful information for aviation forecasters.The data is collected using a Scintec MFAS (Multi-Frequency Flat Array SODAR). While the instrument is capable of a vertical range up to 1000 m, it is optimized for high-resolution monitoring within the lowest 500 m of the troposphere—the most critical zone for aircraft approach and departure. Data is processed at 20-minute intervals, providing a detailed vertical profile of wind direction and speed.The four-year dataset captures the seasonal and diurnal variability of the four dominant local wind regimes:Bora (Bura): A gusty north-easterly wind, often associated with severe mechanical turbulence due to the nearby mountains.Sirocco (Jugo): A moist, south-easterly flow, sometimes with wind speeds above 25m/s in the 500m surface layer.Sea-Breeze (Maestral): A predictable but significant south-westerly coastal circulation that dictates runway changes from standard instrumental approach to a more complex visual approach from north-east, due to mountain proximity on that side of the airport.North-westerly gap flow : Katabatic flow from the mountain gap north-west of the airport, but also a tramontana, typical post-frontal flow (synoptic + orography).Moreover, in aviation meteorology information from SODAR instrument can be very useful for detecting vertical wind shear.This poster will present some typical examples of wind profiles at Split airport, wind roses at different altitudes (e.g., 50m, 150m, 300m), and examples of significant wind shear episodes.These findings emphasize the necessity of ground-based remote sensing for enhancing safety and operational efficiency at topographically challenged coastal airports.
- Open Access
- 14 March 2026
- Published by Copernicus GmbH
Meteotsunamis along the European coastlines: distribution, atmospheric background and generation potential
The work assesses the distribution and strength of meteorological tsunamis (meteotsunamis) along the European coasts. The first step of the assessment is based on the evaluation of historical events observed along the European coasts and on the analysis of decadal time series of 1-minute sea level time series measured at more than 200 tide gauges. The analysis reveals that meteotsunamis are most prominent and contribute most strongly to extreme sea levels along the coasts of the Mediterranean Sea, but also affect other European coasts, including the Black Sea, the Baltic Sea, the North Sea and the European Atlantic coast. The second step of the assessment involves analysing atmospheric conditions during European meteotsunamis. Several atmospheric tsunamigenic sources are suggested: (i) atmospheric gravity waves and convective jumps that occur during otherwise fair weather (so-called “good weather meteotsunamis”), and (ii) convective jumps and pressure changes associated with extratropical cyclones, in particular with their cold fronts. Finally, the potential for the generation of meteotsunamis is assessed by analysing (i) the bathymetric and coastal characteristics of the European coasts and (ii) the prevalence of atmospheric conditions that can generate meteotsunamis. A positive superposition of these two factors leads to the highest meteotsunami generation potential over the Mediterranean Sea, as also suggested by the distribution of historical events and the analysis of 1-minute sea level time series.
- 14 March 2026
- Published by Copernicus GmbH
Open AccessMeteotsunamis along the European coastlines: distribution, atmospheric background and generation potential
The work assesses the distribution and strength of meteorological tsunamis (meteotsunamis) along the European coasts. The first step of the assessment is based on the evaluation of historical events observed along the European coasts and on the analysis of decadal time series of 1-minute sea level time series measured at more than 200 tide gauges. The analysis reveals that meteotsunamis are most prominent and contribute most strongly to extreme sea levels along the coasts of the Mediterranean Sea, but also affect other European coasts, including the Black Sea, the Baltic Sea, the North Sea and the European Atlantic coast. The second step of the assessment involves analysing atmospheric conditions during European meteotsunamis. Several atmospheric tsunamigenic sources are suggested: (i) atmospheric gravity waves and convective jumps that occur during otherwise fair weather (so-called “good weather meteotsunamis”), and (ii) convective jumps and pressure changes associated with extratropical cyclones, in particular with their cold fronts. Finally, the potential for the generation of meteotsunamis is assessed by analysing (i) the bathymetric and coastal characteristics of the European coasts and (ii) the prevalence of atmospheric conditions that can generate meteotsunamis. A positive superposition of these two factors leads to the highest meteotsunami generation potential over the Mediterranean Sea, as also suggested by the distribution of historical events and the analysis of 1-minute sea level time series.