The Adriatic and Ionian seas are part of the Mediterranean Sea positioned between the eastern coastline of Italy, countries of the Balkan Peninsula (from Slovenia to the south through Croatia, Bosnia and Herzegovina, Montenegro and to Albania) and Greece. The Adriatic Sea is an 800 km long, 150 km wide elongated semi-enclosed basin interacting with the Ionian Sea through the Otranto Strait in the southern part (Fig. ). The shallow northern Adriatic Sea is the Mediterranean sub-basin where storm surges reach higher values , mainly triggered by a strong south-easterly moist and warm wind, called sirocco. For this reason, in this area storm surges and waves have been deeply investigated in the past . Tidal dynamics are particularly evident in the northern Adriatic Sea, where the most energetic tidal constituents, the semi-diurnal M2 and the diurnal K1, reach amplitudes of 27 and 18 cm, respectively .

The weather in the Adriatic area is strongly influenced by local orography and small-scale processes . The use of high-resolution meteorological models is essential to capture the temporal and spatial inhomogeneity of north-easterly bora winds, characterised by topographically controlled high-speed wind jets along the eastern shore . The same holds for sirocco: global and regional numerical models have been shown to consistently underestimate its speed due to the fact that orography, and hence the channelling of the air flow, is not well represented at typical model resolution . Long-term analyses of general wind conditions over the Adriatic basin further indicate a trend of reduction of the intensity of wind events – mostly due to bora – and a general increase in terms of frequency, mostly associated with the increasing storminess of sirocco .

The eastern and western coasts of the Adriatic Sea greatly differ in appearance and are therefore differently impacted by sea storms. The western coast is largely sedimentary, with mild sloping and sandy beaches, while the eastern coast is composed of many islands and headlands and is generally high and rocky. Due to its alluvial origin, natural subsidence occurs in the north-western Adriatic Sea because of compaction of fine-grained deposits , which is worsened by the human exploitation of underground water and gas in some areas. Several shallow coastal transitional water bodies are present along the Italian coastline; the main ones are the lagoon of Marano–Grado, the lagoon of Venice, the system of lagoons of the Po delta, the lagoon of Lesina and the lagoon of Varano .

Extreme sea levels cause the flooding of several coastal cities on both sides of the Adriatic Sea , especially when the storm is associated with spring tides . Part of the western coast is below sea level, and therefore it is also very vulnerable to such hazards . These coastal zones are also strongly impacted by north-easterly storms with severe morphological impacts on natural sectors and damage to structures along urbanised zones . Conversely, recurrent meteotsunami events occur on the eastern side of the Adriatic Sea, and particularly on the Croatian coast and islands, causing flooding and damage in some harbours . According to and , the northern Adriatic coastline, due to its low elevation, will be one of the regions in the Mediterranean area most exposed – in terms of coastal risk for flooding and erosion – to future climate change.

A joint asset which could be exploited through fruitful cooperation is the presence in the whole Adriatic–Ionian coastal territories of large networks of sensors and stations. In the Adriatic region, we mapped 35 tide gauges (9 inside the lagoon of Venice) and nine wave stations, with the highest concentration in the northern Adriatic Sea. The location of all reported monitoring stations is illustrated in Fig. , and their general characteristics are summarised in Tables  and for sea level and wave, respectively. The stations' lists are not exhaustive, since there are other monitoring stations active in the area, the data of which were not available at the time of writing this document.

In several cases, the stations are also equipped with sensors for monitoring meteorological (wind speed and direction, sea surface pressure, air temperature, relative humidity, and precipitation) or oceanographic parameters (seawater temperature, salinity, and current speed and direction).

A multi-model ensemble was developed to combine the outcomes of existing ocean and wave forecasting systems, helping with improving the forecast accuracy and reliability on one hand and by adding indications on the forecast uncertainty on the other hand. The error of multi-model ensemble products should be the lowest on average compared to those of the ensemble members . According to , operational forecasts benefit from the combination of different ocean models by considering different physical parameterisations, numerical schemes, model resolutions and forcings.

Several operational ocean forecasting models are currently available for the Adriatic–Ionian region. Here we combined 17 forecasting systems, with 10 predicting sea level height (either storm surge or total water level) and 9 predicting wave characteristics. The general characteristics of the forecasting systems are summarised in Tables  and , respectively, for sea level and wave. MED-Currents and MED-Waves forecasts were retrieved from CMEMS (http://marine.copernicus.eu/). We would like to point out that there are other operational systems active in the area e.g. the pan-European Storm Surge Forecasting System; , the results of which were not available at the time of writing this document.

The different operational models are forced at the surface boundary by several meteorological models (ECMWF, BOLAM, MOLOCH, COSMO, WFR and ALADIN) with a horizontal resolution ranging from 55 to 1.25 km. The length of the ocean forecast is mostly related to the length of the meteorological forecast and varies from 3 to 10 d. There is large variability in the model's set-up in terms of spatial resolution, temporal frequency, spatial domain (Mediterranean Sea, Adriatic Sea and northern Adriatic Sea), grid arrangement (e.g. structured or unstructured) and data format (netCDF and GRIB). Three of the considered systems (Kassandra, MED-Currents and Adriac) account for the current–wave coupling, and two forecasting systems perform data assimilation of tide gauge observations in the operational chain (SIMMb and SIMMe).

TMES is implemented as an internal processing engine which interacts directly with the resource layer to access the datasets (e.g. time series and forecasts) and to deposit the processing results (e.g. ensemble model result, report and statistics). Such outputs are available to the end users and external portal through the common data sharing system and the geoportal web interfaces.

All numerical model results are interpolated, through a distance-weighted average remapping of the nearest neighbours, on a common regular latitude–longitude grid covering the Adriatic–Ionian macro-region with a resolution of 0.02∘. For the purpose of the coastal flooding hazard, the total sea level height must be forecasted. Therefore, the astronomical tide level values obtained by a specific SHYFEM application over the Mediterranean Sea are added to the residual sea level simulated by the operational systems not accounting for the tide (SHYMED, ISSOS, SIMMb, SIMMe and MFS). These obtained sea level heights simulated by the different models are all referred to the geoid. The spread among the operational simulations is expected to represent a measure of the uncertainty of prediction and should be linked to the forecast error so that cases with the largest spread are those with the highest uncertainty and where a large error of the ensemble mean (and also of the deterministic forecast) is more likely . TMES produces results in terms of the ensemble mean and standard deviation, accounting for a measure of the forecast uncertainty .

The vulnerability to sea storms of a particular segment of coast depends on a wide number of variables not only related to the magnitude of the storm but also including the land characteristics and the social and economic activities that distinguish that area. In order to draw a map showing the coastal areas affected by stormy conditions along the Adriatic–Ionian region, the coast is subdivided into segments of variable length in function of morphology, human settlements and administrative boundaries. The coastal assessment units were selected according to the Mediterranean coastal database (MCD) developed by . The MCD segments have an average length of 4.5 km. For each of these units, the database provides information on the characteristics of the natural and socio-economic subsystems, such as vertical land movement, coastal slope, coastal material, and the number of people exposed to sea level rise and to extreme sea levels.

At each location, three sea condition scenarios are computed considering the mean and standard deviation of predicted sea level and wave ensembles:

MIN: ensemble mean - ensemble SD;

Over the whole Adriatic–Ionian coastal region, the nearshore forecasts provided by the TMES were combined with the coastal characteristics (coast material and slope) provided by the MCD database for computing the total water level (TWL). For the coastal segments characterised by sandy beaches, the TWL was computed by combining the sea level height, wave set-up and wave run-up according to Stockdon's formula (R2, the 2 % exceedance level of run-up maxima; ). For gravel beaches and rocky cliffs, other methodologies should be used for estimating wave run-up , but they could not be applied in this study due to the lack of the required detailed coastal information (sediment grain size, type of rocks, and permeability of the structure).

It is well known that the estimation of the total water level is strongly influenced by the local coast typology and morphology and that the MCD segments are sometimes too coarse to represent complex morphologies, especially in confined coastal systems (lagoons) and along the eastern rocky coast. Therefore, in order to provide more reliable and resolute hazard assessment at a very fine coastal scale, the IWS has been designed to provide multi-model forecast products to existing early warning systems, developed in areas were a deep knowledge of the coastal dynamics and high-resolution datasets (topography and bathymetry) are available. In this study, we present three existing local forecasting and early warning systems operative in the Adriatic Sea (Slovenia, the Emilia-Romagna region and the city of Venice), to which IWS provides the information required for issuing coastal risk alerts.

On 29 October 2018, an exceptional storm hit the central and northern part of Italy with very strong south-easterly winds (called sirocco) over the Adriatic Sea. The basic meteorological situation of the 2018 storm was similar to the 1966 and 1979 ones, although with a weaker pressure gradient over the Adriatic area . The weather condition was governed by a semi-stationary upper-level trough which was positioned over the western Mediterranean on 28 October and was only slowly moving north-eastward on 29 and 30 October (Fig. ). The upper-level southerly flow on the eastern side of the trough was very intense due to strong pressure gradients throughout the whole period of the event. The occurrence of the upper-level trough resulted in a formation of a very intense low-level low-pressure system over the Alps and central Mediterranean which was the most prominent surface feature of the event.

The air mass over the Italian Peninsula and Adriatic was very unstable on 28 and 29 October due to the meridional flow which was bringing moist and warm air from North Africa and central Mediterranean. In this sense, it was a typical Autumn situation when the amount of precipitation can be extreme, especially on the windward side of orographic barriers. The flow at the surface was further intensified by extreme convection over the Apennines and the Alps. The amount of precipitation in northern Italy and wind storms over the Alps and northern Adriatic was rather extreme and not often observed in such intensity.

It is worth mentioning that the sirocco wind started already on 26 October at the most of Adriatic and lasted for almost 4 d without interruption, with the strongest wind in the northern Adriatic on 29 October afternoon, just before the passage of the cold front. Most of this time, the sirocco was well developed over the entire Adriatic basin and even further south in the Ionian Sea. This meant that the fetch was exceptionally long for the locations in the northern Adriatic Sea.

Consequently, the sea level rose in the northern part of the Adriatic Sea – the area most affected by flooding – reaching peak values around 13:00 UTC in Venice, Koper and Rovinj (Fig. a). Exceptional sea levels were reached also along the Emilia-Romagna region, with values higher than 1 m for about 5 h, as measured at Porto Garibaldi. It has to be noted that these maximum values were not registered during the storm peak (happening at around 19:00 UTC in this location) due to being out of phase with the astronomical tide. A secondary maximum was recoded in Koper and Rovinj just after the cold front passed and when the wind and waves were decreasing but the tide was rising. Along the central and southern Croatian coast, sea level resulted in being marginally affected by storm surge even if weak meteotsunamis were observed during the frontal passage late in the evening on 29 October.

The very long wind fetch allowed waves to develop over the whole basin. Available wave monitoring stations recorded values of significant wave height (the average height of the highest one-third of all waves measured; SWH) up to 6 m at the Piattaforma Acqua Alta (PTF), 15 km off the shore of the Venetian littoral and up to 4.7 m (8 m of maximum wave height) near the city of Rovinj (Fig. b). Along the north-western Italian coastline, due to its mild slope, wave breaking occurs and the measured SWH reaches values of about 2 m during the storm peak (Nausicaa and Senigallia monitoring stations). On the Gulf of Trieste, the highest waves occurred 6 h later (Zarja wave buoy), probably due to the eastward shift of the wind induced by the passage of the cold front. In the south-eastern Adriatic Sea, high wind and wave values were recorded even before the cold front on 28 October. The highest waves recorded in Dubrovnik reached values of about 5 and 9 m for significant and maximum wave height, respectively. Rough sea conditions (SWH >2.5 m) lasted for 57 h, while the very rough sea state (SWH >4 m) was recorded for 9.5 h. According to long-term time series of observations, the 29 October 2018 event reached the records of the second highest sea state ever measured all along the Adriatic coast.

Here we present the results of the forecasting system at the hourly time step and for the day of the event only. However, as described by , up to 5 d (6 d for the surge) earlier there were indications of a severe event. Figure  shows the TMES results in terms of the ensemble mean and standard deviation for both the sea level height (Fig. a and b) and the significant wave height (Fig. c and d). The storm surge during the 29 October event affected mostly the northern Adriatic Sea (Fig. a), while severe sea conditions occurred over most of the Adriatic Sea, with the higher waves impacting the Croatian coast (Fig. c). The ensemble operational system also provides the estimation of the uncertainty associated with the forecast of this event. Uncertainty is generally higher where the sea level and the waves reach the highest values (Fig. b and d). The ensemble standard deviation reached maximum values of about 30 cm for the sea level and 1.5 m for the significant wave height.

Model forecasts could be extracted at any location in the domain to provide a clear representation of sea storm evolution. As an example, we reported in Fig.  the values extracted at PTF, Rovinj and Dubrovnik (see Fig.  for their location). The comparison with the observations highlights the good performance of the ensemble methodology in reproducing such a strong event. The ensemble mean time-series are smoother than the observations and slightly underestimate the maximum sea level in the northern Adriatic Sea (Figs. 5a and 6b) as well as the peak wave height at 20:00 UTC (5 m of forecasted significant wave height with respect to almost 6 m of observed at PTF; Fig. 6c). However, the observed values are – almost always – within the TMES spread (i.e. the standard deviation of the ensemble members). It is worth noting that the forecast uncertainty increases with the forecast lead time and with the severity of the storm, the maximum of which was reached in the northern Adriatic Sea between 19:00 and 20:00 UTC. In the southern Adriatic Sea (Fig. d), the ensemble mean reproduces the observed significant wave height well, which remained between 3 and 5 m for the whole day. For this specific location the spread of the ensemble assumed values between 0.7 and 1.1 m on 29 October.

In order to provide the perception of the physical processes acting along the Adriatic–Ionian coastal areas that are responsible for storm-related hazards, the results of the multi-model ensemble – in terms of sea level and wave conditions – were processed for each coastal assessment unit. Considering the general underestimation of the ensemble means during the peak of the storm, we will mostly focus our basin-wide storm analysis on the MAX sea condition scenario.

The 2 % exceedance level of wave run-up maxima (R2) computed according the Stockdon's formula and the total water levels forecasted for the 29 October 2018 event (at 19:00 UTC) are reported in Fig.  for the MAX scenario. As for sea level height results (Fig. a), the maximum values of TWL are found in the northern Adriatic along the Veneto and Friuli Venezia Giulia coasts. Indeed, during the 29 October storm, several coastal lowlands in the northern Adriatic were flooded. At these locations, characterised by gently sloping sandy beaches, the estimated R2 reached maximum values of 0.8 m, accounting therefore for about 50 % of the total water level. The combination of the sea level height and the wave set-up and run-up generated high values of the total water level (TWL >1 m) also at some locations on the along the eastern Adriatic coast and the Marche region. Along the Croatian coast, the extremely high waves and the high sea levels caused inundation and widespread damage to the coastal infrastructure (Istria and Zadar). Moreover, because of the rough sea conditions, there was a disruption of the maritime traffic during 27–30 October, and the ferry service cancelled almost all the scheduled sailings on 29 October, so most of the Croatian islands were cut off from the mainland for more than a day. TWL differences between the MEAN and MAX scenarios (not shown) reached the maximum values of about 0.4 m there, which is higher than the standard deviation of the multi-model ensemble for the sea level height.

As stated in Sect. , previous studies demonstrated that the wave run-up estimation increases with the beach slope. Therefore, the high wave run-up values found at some coastal segments (e.g. along the Marche and Abruzzo regions) are due not only to the severe wave conditions but also to the fact that they are characterised by a steep coast (slope >0.15). In such reflective conditions, the use of an alongshore-averaged beach slope in practical applications of the run-up parameterisation may result in large run-up error .

In the following paragraphs, we describe the results of the application of the multi-model ensemble to the existing early warning systems and investigate the details of the storm impact and hazard at the three selected locations.

Due to its northward orientation, the Slovenian coast is relatively well protected from the waves generated by southerly winds, as in the case of the 29 October 2018 storm. Indeed, over there and for this event, the wave contribution to the total water level is negligible. According to the 10 min observation data presented in Fig. , the sea level in Koper reached peak values well above the orange alert level (343 cm at 12:50 UTC and 341 cm at 23:20 UTC) and lasted for almost 10 h above the yellow alert level. As a consequence, the sea flooded several coastal locations, where the firemen set up anti-flooding barrages. As shown in Fig. , the MEAN scenario predicted the exceedance of the yellow flooding alert level but underestimated the observed peak values. A more realistic – even if slightly overestimated – prediction of the flooding hazard for the Slovenian coast is provided by the MAX scenario.

Throughout the Emilia-Romagna region, several coastal sites were affected by flooding and erosion during the 29 October 2018 sea storm due to the combination of the high sea level and energetic wave conditions. The documented coastal impacts are reported in Fig. b and include the erosion of the beach and of the winter dunes, coastal flooding, and damage to coastal infrastructure and defence structures. Damage and impacts were reported especially for the northern part of the region, while along the southern coastal area between Cesena and Riccione, real impact data are not available. The hazard index computed for this event using the XBeach model forced with the three (MEAN, MIN and MAX) conditions reveals that the most severe sea condition scenario (MAX scenario) provides an exceedance of the predefined alert thresholds, indicating a high level of coastal risk. An example of the safe corridor width (described in Sect. ) calculated for a single cross-shore section, located at Lido di Classe, is reported in Fig. . The predicted coastal hazard (Fig. a) shows that the most critical scenario is in good agreement with the documented coastal impacts, displayed in the right panel. For this event, by comparing hazard forecasts and observations, the use of IWS provides a good prediction (MAX scenario) of coastal impacts for the major part of the Emilia-Romagna coastal area. Moreover, considering the distance between the MIN and the MAX conditions, IWS provides useful information about the range of the possible impacts.

On 29 October 2018, the city of Venice was inundated by the exceptional sea storm. At 13:40 UTC the sea level reached the peak value of 156 cm at Punta della Salute (fourth historical level of flooding in Venice since 1872), which put three-quarters of the pedestrian public area of the historic town under water. Sea levels reaching 120 cm (above local datum), at which point flooding covers 28 % of Venice, lasted for about 14 h. The SHYFEM application to the lagoon of Venice, forced by the open-sea TMES results, forecasted sea level peak values of 142 and 161 cm for the MEAN and MAX scenarios, respectively. Figure  shows the corresponding flooding map of the city of Venice according to the predicted peak values (rounded at 140 and 160 cm); 59 % and 77 % of the pedestrian surface is flooded for the two scenarios, respectively. In the first case, the public navigation transport is strongly limited, and only links to the islands are guaranteed; besides this, most of the elevated walkways are unusable. In the second case, the public navigation transport is no longer guaranteed as well as all of the elevated walkways. Moreover, many internal channels are no longer navigable due to insufficient vertical space under some bridges, and hence the emergency rescue response may be strongly delayed. Since the observed peak was 156 cm, the MAX scenario provided a realistic prediction of the flooding hazard for the city of Venice.