Pseudo-global warming projections of extreme wave storms in complex coastal regions: the case of the Adriatic Sea

Abstract

This numerical work aims to better understand the behavior of extreme Adriatic Sea wave storms under projected climate change. In this spirit, 36 characteristic events—22 bora and 14 sirocco storms occurring between 1979 and 2019, were selected and ran in evaluation mode in order to estimate the skill of the kilometer-scale Adriatic Sea and Coast (AdriSC) modelling suite used in this study and to provide baseline conditions for the climate change impact. The pseudo-global warming (PGW) methodology—which imposes an additional climatological change to the forcing used in the evaluation simulations, was implemented, for the very first time, for a coupled ocean–wave–atmosphere model and used to assess the behavior of the selected storms under Representative Concentration Pathway (RCP) 4.5 and RCP 8.5 greenhouse gas projections. The findings of this experiment are that, on the one hand, the AdriSC model is found capable of reproducing both the Adriatic waves associated with the 36 storms and the northern Adriatic surges occurring during the sirocco events and, on the other hand, the significant wave heights and peak periods are likely to decrease during all future extreme events but most particularly during bora storms. The northern Adriatic storm surges are in consequence also likely to decrease during sirocco events. As it was previously demonstrated that the Adriatic extreme wind-wave events are likely to be less intense in a future warmer climate, this study also proved the validity of applying the PGW methodology to coupled ocean–wave–atmosphere models at the coastal and nearshore scales.

1 Introduction

In the past decade, atmospheric regional climate projections—with typical resolutions of 12 to 50-km, have been used in the Mediterranean Sea and most particularly in the Adriatic Sea (Fig. 1a) in a wide range of impact studies, such as the assessment of future wind (e.g. Bellafiore et al. 2012; Belušić Vozila et al. 2019), wave (e.g. Lionello et al. 2012a; Benetazzo et al. 2012; Bonaldo et al. 2017) and storm surge (e.g. Lionello et al. 2012b; Androulidakis et al. 2015; Mel et al. 2013) climates or the characterization of climate-related hazards (i.e. flooding, extreme wave conditions, coastal vulnerability and erosion processes) in the northern Adriatic (e.g. Rizzi et al. 2017; Torresan et al. 2019). In particular, Benetazzo et al. (2012) demonstrated that, in the Adriatic Sea, the observed seasonal wave characteristics could be numerically reproduced at the regional scale and the intensity of the strongest winds (i.e. the so-called bora and sirocco events; Brzovíć and Strelec Mahović 1999; Grisogono and Belušić 2009), and in consequence of the waves, was likely to decrease by the end of the twenty-first century. The study also highlighted that additional research should focus on distinguishing the effect of climate change on extreme bora and sirocco events. which lead to respectively strong air–sea interactions (e.g. Pullen et al. 2006; Janeković et al. 2014; Ličer et al. 2016) impacting the Adriatic thermohaline circulation (Vilibić et al. 2013) and high waves potentially associated with storm surges in the northern Adriatic (Vilibić et al. 2017; Bajo et al. 2019)—more particularly flooding in the Venice Lagoon (Trigo and Davies 2002; Cavaleri et al. 2010), or capable of moving large boulders (weighting up to a ton) along the Croatian coastline (Biolchi et al. 2019a, b).

Fig. 1

a Location of the different Adriatic Sea wave measurements along the Italian and Croatian coastline and b selected 36 one-day extreme wave events during the 1979–2019 period—depending on the wind conditions (bora or sirocco) schematized by arrows in the map, used for the SWAN model evaluation. The notation + 1 (b) means that the 01/01/2015 event is also selected

However, due to the complex orography of the elongated semi-enclosed Adriatic basin surrounded by mountains and associated with bathymetries evolving from a really shallow and wide shelf (300 km in length with less than 80 m in depth) in the north to a deep pit (about 1200 m depth) in the south (Fig. 1a), the evolution of the extreme bora and sirocco wind patterns, and their impact on extreme waves and storm surges, can only be achieved via high-resolution limited-area atmospheric models (e.g. Pasarić et al. 2007; Klaić et al. 2009; Prtenjak and Belušić 2009; Prtenjak et al. 2010; Ricchi et al. 2016; Cavaleri et al. 2018) forcing sea-state and surge models (e.g. Cavaleri et al. 2010, 2019). Additionally, in the recent years, the use of very high-resolution (i.e. kilometer-scale, also known as convection-permitting) regional climate models in atmospheric studies—in particular via the so-called pseudo-global warming (PGW) downscaling method (Schär et al. 1996; Rasmussen et al. 2011), has been proven to greatly improve the future projection of precipitations and convective storms (Pan et al. 2011; Kendon et al. 2014; Tolle et al. 2014; Argueso et al. 2014; Rasmussen et al. 2014; Ban et al. 2014; Prein et al. 2015; Fosser et al. 2016; Kendon et al. 2017). Though, because of their tremendous computational costs, such high-resolution applications have not yet been developed for coupled ocean–atmosphere models.

The present paper thus aims at both (1) implementing and testing the PGW methodology—already used in kilometer-scale atmospheric climate studies, for a complex kilometer-scale atmosphere-wave-ocean modelling suite and (2) quantifying the changes of the Adriatic wave extremes, and their associated storm surges during sirocco events, between present-day (for the 1979–2019 period) conditions and future climate projections (for the 2060–2100 period) for two greenhouse gas scenarios: Representative Concentration Pathway 4.5 (RCP 4.5) and Representative Concentration Pathway 8.5 (RCP 8.5). To this purpose, Sect. 2 first describes in detail the newly developed Adriatic Sea and Coast (AdriSC) modelling suite (Denamiel et al. 2019) used in this study, the set of sirocco and bora storms as well as the wave and sea-level measurements selected in the Adriatic Sea for the 1979–2019 period and, finally, the implementation of the PGW methodology for the ocean–wave–atmosphere AdriSC model. Then, the evaluation of the AdriSC model and the statistical analysis of the climate change impact on the Adriatic extreme waves and storm surges under RCP 4.5 and RCP 8.5 scenarios are presented in Sect. 3. Finally, the validity and limitations of the methodology and results derived in this numerical experiment are discussed in Sect. 4 and some conclusions about the feasibility of using such an approach for long-term climate studies are presented.

2 Model, data and methods

2.1 The AdriSC modelling suite

The Adriatic Sea and Coast (AdriSC) modelling suite (Denamiel et al. 2019) has been recently developed with the aim to accurately represent the processes driving the atmospheric and oceanic circulation at different temporal and spatial scales over the Adriatic and northern Ionian Sea. In this spirit, the AdriSC modelling suite is based on two different modules: a basic module which provides atmospheric and oceanic baroclinic circulation at the deep sea and coastal scales, and a dedicated nearshore module which is used to better reproduce atmospherically-driven extreme events.

The basic module of the AdriSC suite rely on the use and development of the Coupled Ocean–Atmosphere–Wave–Sediment Transport (COAWST) modelling system (Warner et al. 2010). It is built around the Model Coupling Toolkit (MCT) which exchanges data fields and dynamically couples the Weather Research and Forecasting (WRF) atmospheric model, the Regional Ocean Modeling System (ROMS) and the Simulating WAves Nearshore (SWAN) model. The basic module (Table 1) is set-up with two different nested grids of 15-km and 3-km resolution used in the WRF model and covering respectively the central Mediterranean area and the Adriatic-Ionian region, as well as two different nested grids of 3-km and 1-km resolution used for both ROMS and SWAN models and covering respectively the Adriatic-Ionian region (similarly to the WRF 3-km grid) and the Adriatic Sea only.

Table 1 Summary of the AdriSC modelling suite main features

In the nearshore module, the fully coupled ADCIRC-SWAN unstructured model (Dietrich et al. 2012)—covering the entire Adriatic Sea with resolutions ranging from 5-km in the deepest part of the domain to 10 m at the coast (Table 1), is forced every minute with the off-line atmospheric results of a dedicated high-resolution WRF 1.5-km grid. In more details, the hourly results from the WRF 3-km grid obtained with the basic module are first downscaled to a WRF 1.5-km grid covering the Adriatic Sea and the hourly sea surface elevation from the ROMS 1-km grid, the 10-min spectral wave results from the SWAN 1-km grid and finally the 1-min results from the WRF 1.5-km grid are then used to force the unstructured mesh of the ADCIRC-SWAN model.

The AdriSC modelling suite is installed and fully tested on the European Centre for Middle-range Weather Forecast (ECMWF) high-performance computing facilities. Table 1 provides a summary of the AdriSC set-up while more details can be found in Denamiel et al. (2019).

In this study, in order to reproduce the strongest historical wave storms which took place in the Adriatic Sea during the 1979–2019 period and to assess their behavior under climate change projections (RCP 4.5 and RCP 8.5 scenarios), the SWAN model—originally unused in the AdriSC modelling suite, was set-up in both modules to be coupled with the ocean and atmosphere models (i.e. WRF, ROMS, ADCIRC). In the actual configuration, the third generation SWAN model is used with backward space and time propagation schemes, default initial condition, dissipation from whitecapping by Komen et al. (1984) and Madsen bottom friction (Madsen et al. 1988). The wave model receives forcing from WRF 3-km (wind fields) and ROMS 3-km/1-km (ocean surface currents, sea-level and friction) every 10 min in the basic module and from WRF 1.5-km (wind fields) and ADCIRC (ocean barotropic currents, sea-level and friction) every minute in the nearshore module. In addition, the computation of the bottom stress of the ocean models (respectively ROMS and ADCIRC) was updated in order to account for the spatial distribution of the sediment grain size at the bottom of the Adriatic Sea extracted from the Adriatic Seabed database (Jenkins et al. 2005) and the wave effects. For the evaluation runs, during the 1979–2019 period, in order to reproduce the historical storms as accurately as possible, the basic module was set-up to run for 3 days. Initial conditions and boundary forcing were provided the 6-hourly ERA-Interim re-analysis fields (Dee et al. 2011; Balsamo et al. 2015), either the monthly or the daily re-analysis MEDSEA-Ocean fields (Pinardi et al. 2003), depending on whether the storm took place before or after the 1st of January 1987, and either the 6-hourly ERA-Interim wave fields or the hourly MEDSEA-Wave fields (Ravdas et al. 2018), depending on whether the storms took place before or after the 1st January 2006. The nearshore module, forced by the results of the basic module, was set-up to run for the last day and half of the basic module simulations.

As an in-depth sensitivity study of the impact of model resolution on the wave and storm surge representation is out of scope in this work, the advantages of using kilometer scale model in the atmosphere and unstructured meshes in the ocean is briefly discussed in Appendix 1. However, hereafter, only the last 24-h 1-min wave and sea-level results of the nearshore module extracted from respectively the unstructured SWAN model (referred as AdriSC unSWAN in this study) and the ADCIRC model are analyzed.

2.2 Documented historical extreme wave storms in the Adriatic Sea

In the Adriatic Sea, only two most frequent winds—bora and sirocco (Fig. 1a), can produce fetches large enough to drive extreme wave storms (Pomaro et al. 2017). The bora is a cold east-northeast wind which flows through eastern Adriatic mountain passes, being particularly severe along the Croatian coastline where its intensity sometimes surpasses 30 m s⁻¹ and its gust reaches up to 70 m s⁻¹ (e.g. Jiang and Doyle 2005; Kuzmić et al. 2005; Belušić and Klaić 2006; Gohm et al. 2008; Grisogono and Belušić 2009; Trošić 2015). Continuous gale force (> 15 m s⁻¹) bora winds are most common during the cold season (November through March) and have an average duration of 12 h with rare events that can last up to 2 days. The sirocco is a warm southeast wind originating from North Africa, blowing over the Mediterranean Sea and sometimes affecting the Adriatic Sea, being channelized by the surrounding mountains, with gust reaching more than 30 m s⁻¹ (e.g. Poje 1992; Jurčec et al. 1996; Penzar et al. 2001; Pasarić and Orlić 2004). Although sirocco winds are not as strong as the bora, continuous gale force events occur more frequently between October and March, usually lasting 10–12 h—with rare occurrences as long as 36 h, and often bring rain—sometimes mixed with Saharan dust (Cushman-Roisin et al. 2001). Both sirocco and bora episodes may vary in intensity and spatial coverage, extending either over the whole Adriatic or just a part or being conjoined, with sirocco blowing in the southern and bora in the northern Adriatic. Finally, in terms of extreme conditions, as sirocco winds can produce extended fetch, contrarily to the bora winds which are fetch-limited, the largest wave heights were recorded in the northern Adriatic during extreme sirocco events (Leder et al. 1998; Bertotti et al. 2011; Pomaro et al. 2017). These waves can be associated with extreme storm surges in the Venice Lagoon, the Gulf of Trieste (Fig. 1a) and the whole northern Adriatic (Lionello et al. 2012a; Međugorac et al. 2015).

In this study, in order to perform the evaluation of the AdriSC nearshore module, the choice of the studied extreme events was mostly driven by the available information and measurements recorded during the 1979–2019 period. For the sirocco events, the 14 selected storms (Fig. 1b) were extracted from the long-term record of the Venice extreme flooding (https://www.comune.venezia.it/it/content/le-acque-alte-eccezionali). For the bora events, only 22 of the most recent extreme storms were selected (Fig. 1b) as more wave measurements became available in the Adriatic Sea at the end of the twentieth century (more details on the selected bora events are provided in Appendix 2). The majority of the selected bora events peaked in the northern Adriatic, where bora wind is the strongest (Grisogono and Belušić 2009). The set of wave measurements (Table 2)—used to evaluate the skills of the AdriSC nearshore module to reproduce the 36 selected storms, spans between 1979 and 2019 and consists in 6 stations along the Italian coast—Acqua Alta tower from Pomaro et al. (2018), Venice 1 (only for the significant wave height measurements) from the Copernicus Marine Environment Monitoring Service (ftp://my.cmems-du.eu/Core/INSITU_GLO_WAVE_REP_OBSERVATIONS_013_045/history/mooring/) and Venice 2, Ortona, Ancona, Monopoli from the Italian Data Buoy Network managed by ISPRA (Bencivenga et al. 2012), 4 stations along the Croatian coastline—Rovinj, Split, Ploče and Dubrovnik, from the Croatian Hydrographic Institute (Hrvatski hidrografski institut—HHI), and one station in the middle of the northern Adriatic shelf—IVANA-A also from HHI. However, it should be noticed that storm coverage is about 5 times higher from the Italian than the Croatian measurements (Table 2). In addition, concerning the extreme storm surges associated with the sirocco events, two long-term hourly sea-level measurements extracted between 1979 and 2019 from tide gauges located respectively in the Venice Lagoon (at Punta Della Salute, 45.4310° N and 12.3364° E, maintained by ISMAR Venezia) and in the Gulf of Trieste (at 45.6544° N and 13.7561° E, maintained by ISMAR Trieste) were also used to evaluate the skill of the AdriSC nearshore module.

Table 2 Adriatic Sea wave buoy data available along the Italian and Croatian coastlines between 1979 and 2019

2.3 Pseudo-global warming methodology

The two major challenges posed by performing kilometer-scale climate projection simulations are, on the one hand, the relative slowness of the AdriSC modelling suite (a month of results produced per day with the basic module alone), and on the other hand, the low temporal and spatial resolutions (only few vertical levels for daily or monthly data) of the coupled regional climate model (RCM) results available to provide boundary conditions to the WRF 15-km and ROMS 3-km models. To address these concerns, the projection of the extreme Adriatic Sea wave events for the RCP 4.5 and RCP 8.5 scenarios is performed, in this study, via a pseudo-global warming (PGW) method. The principle of the PGW simulations—as first introduced by Schär et al. (1996) and described in details by Rasmussen et al. (2011), Kröner et al. (2017) and Brogli et al. (2019a, b), is to impose an additional climatological change (e.g. a temperature change $\Delta T$ representative of the increase in temperature between past and future climate) to the forcing used to produce the evaluation runs.

In the Mediterranean Sea, one of the specific aim of the Med-CORDEX experiment (https://www.medcordex.eu/)—part of the Coordinated Regional Climate Downscaling Experiment (CORDEX) initiative (https://esg-dn1.nsc.liu.se/search/cordex/) which coordinates the production of climate change projections at the regional scale (Giorgi et al. 2009; Giorgi and Gutowski 2015), is to provide coupled ocean–atmosphere regional model results. The RCMs of the Med-CORDEX ensemble are based on several numerical models running in coupled or uncoupled mode and forced by different Global Climate Models (GCMs). However, at the time of this study, due to a reported issue with the CNRM-CM5 CMIP5 GCM forcing for the historical run (that removes reliability of this product, https://www.medcordex.eu/warnings/Communication-Issue-Files_CNRM-CM5_historical_6hLev_en.pdf), the only coupled results publicly available—with high enough temporal and spatial resolutions for the historical period (1950–2005) and the two climate scenarios RCP 4.5 and RCP 8.5 (2006–2100), were those of the LMDZ4-NEMOMED8 RCM model (Hourdin et al. 2006; Beuvier et al. 2010) forced by the IPSL-CM5A-MR GCM model (simulations r1i1p1). These results—defined as two continuous LMDZ4-NEMOMED8 simulations (1950–2100) extending the historical run with either the RCP 4.5 or the RCP 8.5 runs, are referred as SCEN 4.5 and SCEN 8.5, respectively, and used hereafter to force the PGW simulations. The PGW climatological changes derived from SCEN 4.5 and SCEN 8.5 between the 1979–2019 and the 2060–2100 periods are thus tested in this study.

Finally, the key development of this work is the extension of the PGW method—which had till now only been used in atmospheric models, to the ocean models and more particularly to the AdriSC modelling suite. For the atmosphere, as described in many previous studies (Pan et al. 2011; Kendon et al. 2014; Tolle et al. 2014; Argueso et al. 2014; Rasmussen et al. 2014; Ban et al. 2015; Prein et al. 2015; Fosser et al. 2016; Kendon et al. 2017), the ERA-Interim air temperature $(T^{ERAI} )$, relative humidity $(RH^{ERAI} )$ and horizontal wind velocities ${\mathbf{V}}^{ERAI} = (V_{x}^{ERAI} ,\;V_{y}^{ERAI} )$ defined on 37 atmospheric pressure levels $(p)$ are modified between 1000 and 70 hPa with respectively $\Delta T(t_{{c{ lim }}} ,\;x,\;y,\;p)$, $\Delta RH(t_{clim } ,\;x,\;y,\;p)$ and $\Delta {\mathbf{V}} = (\Delta V_{x} (t_{clim } ,\;x,\;y,\;p),\;\Delta V_{y} (t_{clim } ,\;x,\;y,\;p))$ derived from SCEN 4.5 and SCEN 8.5 by subtracting the atmospheric results from the 1979–2019 period to those of the 2060–2100 period and producing 6-hourly three-dimensional climatologic changes for the 366 days of the year $(t_{clim } )$. The WRF 15-km boundary and initial conditions of the PGW simulations ($T^{SCEN}$, $RH^{SCEN}$, $V_{x}^{SCEN}$ and $V_{y}^{SCEN}$) are thus given by:

$$\begin{aligned} T^{SCEN} (t,\;x,\;y,\;p) &= T^{ERAI} (t,\;x,\;y,\;p) + \Delta T(t_{clim } ,\;x,\;y,\;p) \hfill \\ RH^{SCEN} (t,\;x,\;y,\;p) &= RH^{ERAI} (t,\;x,\;y,\;p) + \Delta RH(t_{clim } ,\;x,\;y,\;p) \\ V_{x}^{SCEN} (t,\;x,\;y,\;p) &= V_{x}^{ERAI} (t,\;x,\;y,\;\;p) + \Delta V_{x} (t_{clim } ,\;x,\;y,\;p) \\ V_{y}^{SCEN} (t,\;x,\;y,\;p) &= V_{y}^{ERAI} (t,\;x,\;y,\;p) + \Delta V_{y} (t_{clim } ,\;x,\;y,\;p) \end{aligned}$$

(1)

In order to adjust the height of the surfaces of constant pressure to the temperature and relative humidity changes, the geopotential—depending on the virtual temperature $T_{v}^{SCEN}$, the ERA-Interim geopotential $\phi^{ERAI}$ at the reference pressure $p_{ref} = 1000\;{\text{hPa}}$ and the gas constant $R$, is recalculated as follow:

$$\phi^{SCEN} (t,\;x,\;y,\;p) = \phi^{ERAI} (t,\;x,\;y,\;p_{ref} ) - \int_{{p_{ref} }}^{p} {\frac{{RT_{v}^{SCEN} }}{p}dp}$$

(2)

Finally, the 2-m air temperature change $\Delta T_{S}$ derived from SCEN 4.5 and SCEN 8.5 runs is used to adjust the ERA-Interim surface (ground and 2-m air) temperatures $(T_{S}^{ERAI} )$ such as:

$$T_{S}^{SCEN} (t,\;x,\;y) = T_{S}^{ERAI} (t,\;x,\;y) + \Delta T_{S} (t_{clim } ,\;x,\;y)$$

(3)

The developed methodology for the ocean follows the principles of the PGW for the atmosphere. In this study, the MEDSEA ocean temperature $(T^{MEDSEA} )$, salinity $(S^{MEDSEA} )$ and currents $(V_{x}^{MEDSEA} ,\;V_{y}^{MEDSEA} )$ defined on 72 unevenly spaced vertical levels $(z)$, are thus modified with respectively $\Delta T(t_{clim } ,\;x,\;y,\;z)$, $\Delta S(t_{clim } ,\;x,\;y,\;\;z)$ and $\Delta {\mathbf{V}} = (\Delta V_{x} (t_{clim } ,\;x,\;y,\;z),\;\;\Delta V_{y} (t_{clim } ,\;x,\;y,\;z))$ derived from SCEN 4.5 and SCEN 8.5 ocean results to produce daily climatologic changes $(t_{clim } )$ for the 366 days of the year. The ROMS 3-km boundary and initial conditions of the PGW simulations ($T^{SCEN}$ and $S^{SCEN}$) are thus given by:

$$\begin{aligned} T^{SCEN} (t,\;x,\;y,\;z) = T^{MEDSEA} (t,\;x,\;y,\;z) + \Delta T (t_{clim } ,\;x,\;y,\;z) \hfill \\ S^{SCEN} (t,\;x,\;y,\;z) = S^{MEDSEA} (t,\;x,\;y,\;z) + \Delta S(t_{clim } ,\;x,\;y,\;z) \hfill \\ V_{x}^{SCEN} (t,\;x,\;y,\;z) = V_{x}^{MEDSEA} (t,\;x,\;y,\;z) + \Delta V_{x} (t_{clim } ,\;x,\;y,\;z) \hfill \\ V_{y}^{SCEN} (t,\;x,\;y,\;z) = V_{y}^{MEDSEA} (t,\;x,\;y,\;z) + \Delta V_{y} (t_{clim } ,\;x,\;y,\;z) \hfill \\ \end{aligned}$$

(4)

In the ocean, the static stability depends on the density $(\rho )$ and the vertical variations of the local potential density $(\sigma_{n} )$ such as:

$$E(x,\;y,\;z) = - \frac{1}{\rho (x,\;y,\;z)}\frac{{\delta \sigma_{n} }}{\delta z}$$

(5)

The stability of the ocean forcing (at the boundaries and for the initial condition) is thus ensured by imposing $E^{SCEN} \ge 0$ at all vertical levels. Finally, the sea surface elevation change $\Delta ssh$ derived from SCEN 4.5 and SCEN 8.5 runs is used to adjust the MEDSEA surface layer $(ssh^{MEDSEA} )$ such as:

$$ssh^{SCEN} (t,\;x,\;y) = ssh^{MEDSEA} (t,\;x,\;y) + \Delta ssh(t_{clim } ,\;x,\;y)$$

(6)

The temperature changes $(\Delta T)$ imposed at the boundaries of both the ocean and atmosphere models are illustrated in Fig. 2. The vertical variations of the spatially- and time- averaged $\Delta T$ presented in Fig. 2b clearly show that, near the surface of the earth, the differences in temperature between scenarios RCP 4.5 and RCP 8.5 reach more than 1.5 °C (for both the ocean and the atmosphere). For the ocean, no significant difference between the two scenarios is seen below depth of 1000 m. For the atmosphere, this difference only starts to decrease above 400 hPa and is minimized above 100 hPa. In addition, the time variations of the spatially-averaged $\Delta T$ for scenario RCP 8.5 (Fig. 2c) highlights that the temperature change imposed to the atmosphere at 2 m height is, most of a year, at least 0.5 °C higher than the one imposed to the sea surface temperature. Finally, Fig. 2d, e present the vertical variations of the temporally-averaged $\Delta T$ along the southern and western boundaries of both the atmosphere and ocean models and illustrate the importance of using spatially varying temperature changes for realistic climate simulations. In Fig. 3, the surface distribution of the temporally-averaged RCP 8.5 changes show that: for the atmosphere, the orography plays a major role in terms of the intensity of the changes (i.e. the strongest increase in temperature, decrease in relative humidity and change in wind speed are generally found at the highest altitudes), and, for the ocean, the changes imposed to the Adriatic and northern Ionian Seas (i.e. strongest increase in temperature and salinity) do not correspond to the changes imposed in the western side of the domain where the strongest changes in current speed occur. Concerning the sea surface elevation, the RCP 8.5 changes are mostly negative and only of the order of a few centimeters (with a maximum of 8 cm). Given that on the one hand, the open boundary of the LMDZ4-NEMOMED8 model (similarly to all the Med-CORDEX simulations, Adloff et al. 2018) does not properly include the projected Atlantic sea-level changes, but just takes into account the thermosteric effects and, on the other hand, the thermal stretching is balanced by the haline shrinking, these results are in accordance with the estimated − 7 cm to 13 cm expected in the Mediterranean Sea (Tsimplis et al. 2008; Jordà and Gomis 2013; Gualdi et al. 2013). Thus, for realistic sea-level projections, mass change-induced sea-level increase—approximated to 50–60 cm in the Mediterranean till 2100 (Jordà and Gomis 2013), should be added to the presented PGW sea-level estimates.

Fig. 2

a Spatial domain and boundaries of the WRF 15-km model and, within the red box, the ROMS 3-km model. b Vertical variations of the spatially- and temporally-averaged temperature changes $\Delta T$ for scenarios RCP 4.5 and RCP 8.5 following pressure level in the atmosphere and depth in the ocean. c Time evolution depending on the day of a year (DOY) of the spatially-averaged 2-m air (in green) and sea-level (in blue) climatologic temperature changes $\Delta T$ for scenario RCP 8.5. Vertical structure of the temporally-averaged temperature changes $\Delta T$ (RCP 8.5) imposed at the southern and western boundaries of d the WRF 15-km model and e the ROMS 3-km model

Fig. 3

Surface distribution of the temporally-averaged RCP 8.5 changes of a temperature ($\Delta T$), relative humidity ($\Delta RH$) and wind speed ($\Delta V$) in the atmosphere for the WRF 15-km domain and b temperature ($\Delta T$), salinity ($\Delta S$) and current speed ($\Delta V$) in the ocean for the ROMS 3-km domain. The variations of the sea surface elevation ($\Delta ssh$) RCP 8.5 changes are presented in c as temporally-averaged surface distributions and time-varying open boundary conditions

In addition to the changes imposed to the ERA-I and MEDSEA forcing presented in the previous paragraphs, the volume mixing ratio of five atmospheric gases (carbon dioxide, methane, nitrous oxide and chlorofluorocarbons 11 and 12) used in the evaluation runs is modified in the scenario runs using projected values (Bernstein et al. 2008) averaged between 2060 and 2100 (Table 3). Further, the historical monthly Adriatic Sea river discharges are climatologically changed for the RCP 4.5 and RCP 8.5 scenarios (Fig. 4) following the study of Macias et al. (2018). Concerning the waves, the forcing used in the evaluation simulations were kept unchanged for the scenario runs as the required data needed to apply the PGW methodology to the waves was not available. However, since the open boundary of the ROMS 3-km grid is located at least 400-km south of the Strait of Otranto, the wave field within the Adriatic basin is not considered to be highly affected by the propagation of these forcing. Finally, as this study aims to estimate the impact of climate change on atmospherically-driven extreme events and not to forecast future storms, the tidal forcing imposed for the evaluation runs was also kept unchanged for the scenario runs.

Table 3 Atmospheric gas volume mixing ratios used for the evaluation and scenario runs

Fig. 4

Monthly climatologic changes (in percentage) imposed to the Adriatic Sea river discharges for scenario runs RCP 4.5 and RCP 8.5

To summarize, the set of 108 runs used in this study consists in 36 Adriatic Sea wave storm simulations (selected in Sect. 2.2) carried out with the AdriSC modelling suite (described in Sect. 2.1)—for 3 days within the general module (i.e. coupled WRF-ROMS-SWAN) and 1.5 day within the nearshore module (i.e. WRF 1.5-km and coupled ADCIRC-unSWAN), in evaluation mode first and then in climate change mode, for both RCP 4.5 and RCP 8.5 scenarios, imposing the PGW methodology (presented in Sect. 2.3).

3 Results

3.1 Evaluation of the AdriSC nearshore module wave component

To assess the skill of the AdriSC unSWAN model, the last 24-h of the 1-min significant wave height, peak wave period and mean wave direction results are extracted from the 36 simulations carried out in evaluation mode at the 11 locations of the wave stations presented in Sect. 2.2. The data are analyzed in three steps (see Figs. 5, 6, 7). First, the overall behavior of the model is presented as a scatter plot (Figs. 5a, 6a, 7a) for the entire set of simulations and measurements. Then, the quantile–quantile distributions of the wave parameters are displayed separately for the Italian and Croatian wave stations (Figs. 5b, 6b, 7b) and, finally, the performance of the unSWAN model wave distributions during bora and sirocco events (Figs. 5c, 6c, 7c) is illustrated with violin plots (Hintze and Nelson 1998).

Fig. 5

Evaluation of the AdriSC unSWAN significant wave height distributions against measurements a for all the available stations and selected storm events as a scatter plot showing the density (number of occurrences #) with hexagonal bins, b separately for the Italian and Croatian stations as a quantile–quantile plot, and c separately for the bora and sirocco events as violin plot distributions

Fig. 6

As in Fig. 5, but for the peak wave period

Fig. 7

As in Fig. 5, but for the mean wave direction

On the whole (Figs. 5, 6, 7), for the 36 studied storm events, the unSWAN model is in good agreement with the available wave measurements (significant height, peak period and mean direction): (1) in the scatter plots, the points with higher density (in red) are mostly located along the reference lines, (2) the quantile–quantile distributions for both the Croatian and Italian stations also follow the reference lines, except for the mean direction which is not well reproduced for the Croatian stations, and (3) the shapes of the violin plots for the unSWAN model results are similar to those obtained for the measurements during both bora and sirocco events.

However, in more detail, some discrepancies between the measurements and the unSWAN model results can be noticed. For the significant wave height (Fig. 5), (1) the spread of the scatter plot (Fig. 5a) increases from about 0.25 m up to 2 m when reaching the highest values, (2) the model slightly overestimates (up to 0.25 m) the values between 2 and 4 m for both the Croatian and Italian stations and considerably overestimates (up to 1 m) the values above 4 m for the Croatian stations (Fig. 5b), and (3) the mean and median of the model distributions (Fig. 5c) are also overestimated for the bora (2.41 m vs. 2.17 m and 2.52 m vs. 2.13 m, respectively) and for the sirocco (2.13 m vs. 2.05 m and 1.94 m vs. 1.67 m, respectively) events.

For the peak wave period (Fig. 6), (1) similarly to the significant wave height, the spread of the scatter plot (Fig. 6a) increases from about 1 s up to 6 s when reaching the highest values (to be noted: the discontinuities seen in the plot result from the fact that some measurements were provided as integer values), (2) the model slightly underestimates (up to 1 s) the values below 5 s and above 9 s for the Italian stations (Fig. 6b), and (3) the mean and median of the model distribution for the bora events (Fig. 6c) are overestimated (6.88 s vs. 6.61 s and 7.67 s vs. 7.10 s, respectively). Finally, for the mean wave direction, the major problem is the very large underestimation (up to 200°) of the values between 0° and 200° for the Croatian stations (Fig. 7b).

In a nutshell, the unSWAN model seems to have more difficulties to represent the wave conditions during bora events than during sirocco events, which means that the WRF 1.5-km model is most probably overestimating the intensity of the bora winds. Further, the model is capable of reproducing the intensity of the extreme wave events (see quantile–quantile distributions) but not their timing (see spread of the scatter plots), and have better agreement with measurements along the Italian coast than along the Croatian coast. For the last point, the analysis of each Croatian station (not shown here) reveals that the mismatching of the model for the wave directions between 0° and 200° principally occurs at the IVANA-A and Dubrovnik locations which, as the model is in good agreement with the data for the other nine stations, likely results from a problem with the measurements at these two locations. Beside these limitations, the evaluation of the unSWAN model has shown that the newly added wave component of the AdriSC modelling suite can be used to reproduce the historical Adriatic wave storms with a good level of accuracy.

3.2 Impact of climate change on Adriatic extreme waves

With the aim to quantify the climate change impact on the Adriatic extreme wave events under both RCP 4.5 and RCP 8.5 projections for the 2060–2100 period, two kind of results are statistically analyzed: the spatial variations of the extreme wave conditions (Figs. 8, 9, 10), and the temporal variations of the wave parameters at chosen locations along the Adriatic Sea (Figs. 11, 12, 13, 14).

Fig. 8

Baseline (present climate) plots defined as the median, over the entire Adriatic Sea, of the maximum significant wave heights, the maximum peak wave periods and the mean wave directions of the 22 bora (top panels) and 14 sirocco (bottom panels) events

Fig. 9

Climate change (RCP 4.5 and RCP 8.5) impact on the waves defined as the median and root-mean-square (RMS) of the climate adjustments (scenario minus evaluation results) of the maximum significant heights, the maximum peak periods and the mean wave directions of the 22 bora events

Fig. 10

As in Fig. 9, but for the 14 sirocco events

Fig. 11

Peak wave period (Tp) vs. significant wave height (Hs) distributions derived from the 1-min AdriSC unSWAN evaluation and future climate projection (RCP 4.5 and RCP 8.5) results of the 22 bora events and extracted at 5 open Adriatic Sea locations (O₁ to O₅). The results are presented as a combination of scatter plots displaying the distributions of the peak period vs. the significant height

Fig. 12

Mean wave direction distributions derived from the 1-min AdriSC unSWAN evaluation and climate projection (RCP 4.5 and RCP 8.5) results of the 22 bora events and presented as rose plots at 5 open Adriatic Sea locations (O₁ to O₅)

Fig. 13

As in Fig. 11, but for the 14 sirocco events

Fig. 14

As in Fig. 12, but for the 14 sirocco events

The spatial analysis of the extreme wave conditions consists first in defining the baseline conditions (Fig. 8), which are presented as the median over the ensembles of the 22 bora and the 14 sirocco storm simulations in evaluation (present climate) mode. The considered parameters are the maximum significant wave heights, maximum peak wave periods and mean wave directions—calculated for each storm over the last 24-h results of the AdriSC nearshore module. Then, the climate change impact on the wave extremes is given by the differences (referred hereafter as climate adjustments) in maximum significant wave heights, maximum peak wave periods and mean wave directions between the climate change simulations (with the RCP 4.5 and RCP 8.5 scenarios treated separately) and the evaluation runs. Finally, the median and root-mean-square (RMS) of these climate adjustments are calculated for the ensembles of the 22 bora (Fig. 9) and the 14 sirocco (Fig. 10) events. The analysis of the baseline conditions (Fig. 8) shows that the typical significant wave heights and peak wave periods are above 3.5 m and 8 s, respectively. For the bora events, this particularly applies to the Italian coast between 42° N and 45° N of latitude, peaking between 44° N and 45° N latitude with the respective values of 5 m and 10 s. This is the result of the maximum in both bora speed and outreach, coming off the Croatian city of Senj at latitude 44.99° N (the Senj Jet, Grisogono and Belušić 2009). The typical wave propagation for the analyzed bora episodes is mostly towards south-west and west in the northern Adriatic and north-westward in the south. The latter indicates that majority of the selected severe bora events peaked in the northern Adriatic, while being mostly conjoined with sirocco conditions in the southern Adriatic. For the sirocco events, wave heights are substantial in the open Adriatic Sea for the entire domain (except close to the Italian shoreline and in the coastal Croatian area), peaking with values up to 6 m associated with wave periods of 10 s between 44° N and 45° N latitude off the Croatian islands and coast. The typical wave propagation for sirocco events is towards north and north-west.

Typical significant wave heights and peak wave periods are foreseen, in the future climate, to overall decrease during bora events (Fig. 9), for both RCP 4.5 and RCP 8.5 scenarios. Negative median of the climate adjustment is projected over the entire Adriatic domain, except at its southern part in the RCP 4.5 scenario. For the RCP 8.5 scenario, the negative median of the climate adjustments reaches up to 1 m in significant height and 1 s in peak period off the Italian coast between 41°N and 43°N latitude and is associated with larger RMS values surpassing 0.8 m and 1 s, respectively. The direction of the bora winds is also affected over the entire domain (up to 50° change of direction for the RCP 8.5 scenario), but mostly along the Italian coastline between 42° N and 43° N latitude and regions where strong wind shear occur in the bora jets.

For the sirocco events (Fig. 10), the typical significant heights and peak periods are also mostly decreased in the northern Adriatic for both RCP 4.5 and RCP 8.5 scenarios (i.e. small or negative median of the climate adjustments) but increased (i.e. positive median of the climate adjustments) for the rest of the domain.

For the RCP 8.5 scenario, the northern Adriatic decreases can reach up to 0.5 m in significant height and 0.8 s in peak period and are associated with large RMS above 0.6 m and 0.5 s, respectively. Furthermore, the direction of the sirocco waves seems to be totally unchanged (both median and RMS of the climate adjustments are low) for both RCP 4.5 and RCP 8.5 scenarios.

The analysis of the temporal variations of the wave parameters during the selected Adriatic wave storms is based on the 1-min unSWAN series extracted at 5 open Adriatic Sea locations (O₁–O₅, Figs. 11, 12, 13, 14). A comparison between the distributions obtained in the evaluation and future climate (for RCP 4.5 and RCP 8.5 scenarios) simulations is performed, separately for the ensemble of the 22 bora and 14 sirocco events. The results are presented as a combination of scatter and probability density function (PDF) plots for the significant wave height and peak wave period parameters (Figs. 11, 12, 13), and as polar histogram plots for the mean wave direction (Figs. 12, 13, 14).

For the bora events (Figs. 11, 12), the scatter plots at the five chosen locations (Fig. 11) reveal that the overall distribution of the peak wave period vs. the significant wave height is not significantly modified under climate change projections and is presenting a strong linear relationship at all locations (i.e. the peak periods tend to increase linearly with the significant wave heights), with a little spread in the northern Adriatic only (locations O₄ and O₅). However, the analysis of the PDF distributions confirms that both peak wave periods and significant wave heights during bora events are likely to decrease under climate change projections:

• concerning the significant wave height distributions, the values are consistently lowered in the future climate, with a minimum of 0.25 m at location O₅ (northern most part of the Adriatic Sea) for the RCP 4.5 scenario and a maximum of 2 m at location O₄ (where strongest bora wind are likely to blow) for the RCP 8.5 scenario while the tail is generally becoming less heavy under RCP 8.5 scenario at locations O₃, O₄ and O₅ (i.e. the probability of significant wave heights above 3 m is largely reduced), but does not significantly change in the southern Adriatic (locations O₁ and O₂), at locations under the sirocco influence, for both RCP 4.5 and RCP 8.5 scenarios;

• concerning the peak wave period distributions, the values are also lowered at locations O₂, O₃, O₄ and O₅, with a minimum of 0.5 s at location O₂ and a maximum of 1.5 s at location O₃, both obtained for the RCP 4.5 scenario while, as for the significant wave height, the tail is generally becoming less heavy under RCP 8.5 scenario at all locations (i.e. the probability of the peak wave periods above 7 s is largely reduced), but does not present major changes for RCP 4.5 scenario, except at location O₅ (where the tail is clearly less heavy), and at location O₁ (where the tail is more heavy).

Regarding the mean wave direction distributions (Fig. 12), the most significant changes appeared at location O₃ where the waves primarily propagated westward in the evaluation mode, while they shift southward and south-eastward for both RCP 4.5 and RCP 8.5 scenarios. This behavior can also be noticed, in a smaller measure, at locations O₂, O₄ and O₅ (Fig. 12), where however the main direction of propagation is unchanged. Finally, at location O₁, the most noticeable changes in direction occur for the RCP 4.5 scenario, where the south-westward waves are shifted to north-westward and south-eastward, but the main direction of propagation (i.e. north north-westward) is also unchanged.

For the sirocco events (Figs. 13, 14), as for the bora events, the scatter plots at the five chosen locations (Fig. 13) reveal that the overall distribution of the peak wave period vs. the significant wave height is not substantially modified under climate change projections. However, these distributions do not present a strong linear relationship in the northern Adriatic and show that, at locations O₃, O₄ and O₅, for peak periods above 7 s, significant wave heights can vary between 1 and 7 m. The analysis of the PDF distributions reveals that, compared to the bora events, less dramatic changes are to be expected concerning the sirocco wave parameters under RCP 4.5 and RCP 8.5 scenarios:

• concerning the significant wave height distributions, the values are mostly unchanged, even though slightly (about 0.25 m in average) increased in the southern Adriatic Sea and decreased in the northern Adriatic Sea while the tail is, however, generally becoming less heavy under RCP 8.5 scenario in the northern Adriatic Sea at locations O₄ and O₅ (i.e. the probability of significant wave heights above 3 m is reduced) but slightly heavier or unchanged for the remaining locations under RCP 8.5 scenario and for all locations under RCP 4.5 scenario;

• concerning the peak wave period distributions, the values are also mostly unchanged at all locations and for both climate change scenarios, but the respective probabilities are increased in the southern Adriatic (locations O₁, O₂ and O₃) while the tail is generally becoming slightly less heavy under RCP 8.5 scenario in the northern Adriatic (i.e. the probability of the peak wave periods above 9 s is reduced) but does not present major changes for RCP 4.5 scenario.

Finally, the mean wave direction distributions (Fig. 14) show that the north-westward main direction of the waves in evaluation mode remains unchanged under both RCP 4.5 and RCP 8.5 scenarios.

To summarize, the spatial variations of the extremes and the 1-min time series—extracted along the open Adriatic Sea, reveal that, under warming climate change (for both RCP 4.5 and RCP 8.5 scenarios), significant wave heights and peak wave periods are likely to, on the one hand, strongly decrease over the entire domain with a south-eastward shift of direction in the central Adriatic during the extreme bora events, and, on the other hand, decrease in the northern Adriatic with no significant change in direction during the extreme sirocco events. These results are in good agreement with the study of Belušić Vozila et al. (2019) who estimated the possible future changes in wind speed over the Adriatic region, for the 2041–2070 period, from an ensemble of 19 high-resolution (0.11°) CORDEX simulations and found that overall the mean wind speed as well as the number of storms is reduced under the RCP 8.5 scenario for both bora and sirocco conditions. However, Belušić Vozila et al. (2019) also highlights an increase of the bora mean wind speed in the northern Adriatic, which is not in accordance with the presented results (Figs. 9, 11). Although selection of bora events differs between the two studies, this result may indicate the limitation of the PGW methodology which can only be used to assess how past storms would behave under climate change. Additionally, the presented results are in good agreement with other wave climate studies, which all envisage a decrease of wave heights in the Adriatic Sea, in particular concerning sirocco events (Benetazzo et al. 2012; Lionello et al. 2012a; Bonaldo et al. 2017; Pomaro et al. 2017).

3.3 Impact of climate change on northern Adriatic storm surges

In terms of the climate change impact on storm surges, the flooding of the coastal cities along the Adriatic coast—and most particularly in the northern Adriatic, is known to be driven by extreme sirocco conditions (e.g. Robinson et al. 1973; Cavaleri 2000; Cavaleri et al. 2010; Raicich 2015; Medugorac et al. 2015), such as those selected in Sect. 2.2 which led to the highest water levels recorded in Venice lagoon between 1979 and 2019. The atmospherically-driven extreme sea-level changes in the northern Adriatic under RCP 4.5 and RCP 8.5 scenarios and following the PGW methodology can thus be assessed in this study with the ensemble of the selected 14 sirocco events.

The AdriSC ADCIRC model capability to reproduce the storm surges in the northern Adriatic and more specifically in the Venice Lagoon and the Gulf of Trieste is first assessed with a sea-level distribution quantile–quantile analysis (Fig. 15a) of the available hourly measurements and the ADCIRC model results—extracted at the locations of the two tide gauges (Sect. 2.2) from the last 24-h results of the 14 sirocco simulations carried out in evaluation mode. To be noted, (1) as the bathymetry used in the ADCIRC model may be imprecise and may use a different vertical reference level than the tide gauges, the local mean sea-levels of the model results at Venice and Trieste locations were adjusted by adding the difference between the measured and the modelled mean values calculated for the ensemble of the 14 sirocco events; (2) as tides play an important role on extreme storm surges, they were not removed from the sea-level signals, and (3) as this study only aims to present the climate change impact on the sea-level distributions during sirocco events in the northern Adriatic Sea and not to reproduce individual storm surges, the model evaluation can be performed with the presented quantile–quantile analysis.

Fig. 15

Analysis of the northern Adriatic storm surge distributions during the 14 sirocco events: a quantile–quantile analysis of the AdriSC ADCIRC results and the measurements at Venice and Trieste tide-gauge stations, b baseline sea-level plot defined as the median of the maximum sea-levels generated by each storm, c, d sea-level distributions derived from the 1-min AdriSC ADCIRC evaluation and climate projection (RCP 4.5 and RCP 8.5) results and extracted respectively at Venice and Trieste tide gauge stations

Similarly to the analysis performed for the waves in Sect. 3.2, the impact of climate change for the 2060–2100 period on the northern Adriatic storm surges under both RCP 4.5 and RCP 8.5 projections, is then estimated via the statistical analysis of two kind of results: the spatial variations of the maximum sea-levels obtained for each storm (Figs. 15b, 16) and the 1-min sea-level temporal variations at both the Venice Lagoon and the Gulf of Trieste tide gauge locations (Fig. 15c, d).

Fig. 16

Climate change (RCP 4.5 and RCP 8.5) impact on the northern Adriatic extreme sea-levels defined as the median and root-mean-square (RMS) of the climate adjustments (scenario minus evaluation results) of the maximum sea-levels for the 14 sirocco events

For the spatial variations, the sea-level baseline condition (Fig. 15b) is defined as the median over the ensemble of the 14 selected storm simulations in evaluation mode and their maximum sea-levels—calculated for each storm over the last 24-h results of the AdriSC nearshore module. As in Sect. 3.2, the climate change impact on the storm surges is given by the differences (referred hereafter as climate adjustments) in maximum sea-levels between the climate change simulations (with the RCP 4.5 and RCP 8.5 scenarios treated separately) and the evaluation runs. The median and root-mean-square (RMS) of these climate adjustments are calculated for the ensemble of 14 sirocco events (Fig. 16). In terms of the results, the baseline condition (Fig. 15b) shows that typical storm surges during sirocco events reach 1.4 m in the Venice Lagoon and 0.9 m in the Gulf of Trieste and, under climate change conditions (Fig. 16), are decreased (i.e. negative median of the climate adjustments) by more than 0.25 m over the entire northern Adriatic domain for both RCP 4.5 and RCP 8.5 scenarios, and by more than 0.35 m in the Venice Lagoon for the RCP 8.5 scenario. The decrease of the storm surges in the future climate is also associated with an important variability (i.e. RMS) of about 0.35 m in average for both RCP 4.5 and RCP 8.5 scenarios and reaching more than 0.45 m in the Venice Lagoon for RCP 8.5 scenario.

These results are confirmed by the temporal analysis of the sea-level distributions (for the evaluation runs and the RCP 4.5 and RCP 8.5 projections) at Venice and Trieste tide gauge locations (Fig. 15c, d). It appears that, at both locations under RCP 4.5 and RCP 8.5 climate change, the modes of the sea-level distributions are decreased by about 0.25 m while the tails of the distributions are less heavy (i.e. the probability of storm surges above 0.75 m is greatly decreased) and the maximum surges are reduced by about 0.25 m. Finally, the probability of sea-levels below -0.5 m (which was the minimum reached in evaluation mode) is largely increased which reveals that, under the climate change projections, some of the strong sirocco events simulated in evaluation mode might take more time to develop (or never developed) as storms of lower intensity in the RCP 4.5 and RCP 8.5 simulations. Thus, for these events, the wind-wave set-up—building up in the northern Adriatic during the 3 days of simulation in evaluation mode, was decreased as it built up over a smaller period with weaker winds (or did not built up at all).

To summarize, under climate change projections with imposed PGW methodology, the northern Adriatic sirocco storm surges are not only likely to decrease but also to be less frequent, which is in accordance with the analysis of the sirocco wind and storm surge projections (Lionello et al. 2012a; Androulidakis et al. 2015; Belušić Vozila et al. 2019). However, as the subsidence of the Venice Lagoon is not considered in this study, and the sea-level change is imposed as a mean effect for the entire 2060–2100 period in the PGW methodology ignoring the Atlantic global sea-level rise (see discussion about sea surface elevation in Sect. 2.3), these results do not imply that flooding of the Venice city will be less likely in the future.

4 Discussion and conclusions

Understanding how climate change could impact extreme wave storms—one of the most devastating natural hazards occurring along the littoral, is of crucial importance for the future of coastal communities. However, in order to properly capture such atmospherically-driven extreme events and their repercussions on the coast (e.g. extreme wave and storm surges, coastal erosion, etc.), the implementation of computationally expansive kilometer-scale coupled ocean–wave–atmosphere models is required for each specific coastal region considered. Consequently, the classical approach used in climate studies (i.e. 30 years of evaluation run, 50 years of historical run and 100 years of scenario runs) is too impractical and costly to be applied for this kind of investigations. As an alternative, the pseudo-global warming (PGW) methodology—originally developed for kilometer-scale atmospheric studies, presents the advantage of generating at a limited computational cost (i.e. once the PGW climatological forcing is created, each event can be simulated over a short period of time) an ensemble of storms used to statistically assess the impact of climate change on extreme events.

The principal novelty of this study was thus to implement the PGW method within the AdriSC modelling suite—a multi-model chain dedicated to the study of the Adriatic Sea (Denamiel et al. 2019), which couples the atmospheric model WRF at 15-km, 3-km and 1.5-km of resolution with the ocean and wave models ROMS and SWAN at 3-km and 1-km of resolution and the unstructured ADCIRC and SWAN models with up to 10 m resolution along the coast. In particular, a lot of attention was paid to the best way to represent not only the ocean forcing (salinity, temperature, currents) but also the sea-levels, the rivers, the waves and the tides. As the Adriatic Sea collects up to a third of the Mediterranean Sea fresh water budget (Ludwig et al. 2009), the river discharge forcing—which is projected to be highly impacted by climate change, was modified following the study of Macias et al. (2018). However, due to the known uncertainties and/or lack of data linked to sea-level rise and wave climate projections over the entire Mediterranean Sea (Ruti et al. 2016; Adloff et al. 2018), these forcing were kept unchanged in this work. Additionally, in order to only account for the impact of global warming in the storm surge analysis, the tidal forcing also remained untouched. If these approximations have little consequences on the extreme wave results (as the boundary of the SWAN 3-km model is far enough from the studied area), they can impact the presented storm surge distributions due to the non-linear nature of their interactions with the nearshore water depths—including local sea-levels and tides (Johns et al. 1985; Speer and Aubrey 1985; Parker 1991; Zhang et al. 2017; Yang et al. 2019). The impact of sea-level rise was thus ignored in this work and only the atmospherically-driven storm surge distributions were analyzed.

The other important component of this numerical work was to provide a thorough evaluation of the AdriSC modelling suite skill to reproduce historical extreme events and to provide meaningful climate projections via the PGW method. To achieve these goals an ensemble of 22 bora and 14 sirocco extreme wind-wave events were selected between 1979 and 2019 and ran in both evaluation and climate projection (for RCP 4.5 and RCP 8.5 scenarios) modes. The evaluation of the distributions of both the wave parameters (significant height, peak period and mean direction) against 11 stations located along the Adriatic coast, and the storm surges against the Venice and Trieste tide gauges, revealed that overall the AdriSC model is capable of reproducing the selected 36 historical extreme events. Concerning the climate simulations with the PGW method, the wave and storm surge distributions—showing a general decrease of the extreme bora and sirocco intensity for both RCP 4.5 and RCP 8.5 scenarios, follow the previous studies published in the Adriatic Sea (Benetazzo et al. 2012; Lionello et al. 2012a; Androulidakis et al. 2015; Bonaldo et al. 2017; Pomaro et al. 2017; Belušić Vozila et al. 2019) and thus the statistical approach consisting in running ensembles of short simulations for extreme events seems to provide robust results. However, it should be noticed that, (1) as only a small ensemble of storms was selected, the simulated wave and storm surge distributions may not be fully representative of neither the historical Adriatic extreme events between 1979 and 2019 nor their future projections for the 2060–2100 period, (2) as the simulations were performed over a three-day period, the last 24-h results analyzed in this study might be influenced by the imposed initial conditions and finally, (3) as the same ensemble of storms is used in evaluation and climate projection modes, the frequency of the extreme events cannot be analyzed with the PGW method. Additionally, due to the lack of reliable ocean–atmosphere Med-CORDEX runs at the time of this study, the climatological forcing used in the presented PGW simulations were derived from a single model instead of an ensemble of Regional Climate Models (RCMs) which would have provide more robust climate change projections. The generation of such a forcing, or potentially of an ensemble of PGW simulations—as already adopted by the atmospheric community (e.g. Li et al. 2019), could be achieved in a near future when more ocean–atmosphere RCMs will become available in the Mediterranean Sea.

Notwithstanding the limitations of the PGW approach implemented in the AdriSC modelling suite, this study demonstrates that such a method has the potential to be applied to kilometer-scale ocean–atmosphere models for long- or short-term simulations which can be embedded in the traditional CORDEX sub-domains including an oceanic component (e.g. Ruti et al. 2016; Tinker et al. 2016; Zou and Zhou 2016; Han et al. 2019). For coastal areas such as the Adriatic Sea, this can open the door to a better understanding of the climate change impacts on various processes (e.g. dense water formation, ocean sea surface thermal interactions during storms and hurricanes, etc.) closely depending on the air–sea feedback mechanism.

Appendices

Appendix 1

Despite an in-depth sensitivity study of the model resolution impact on extreme event representation being out of scope of this paper, the reasons why kilometer-scale resolution in the atmosphere and meter-scale resolution in the ocean are a pre-requisite to meaningful wave storm modelling in the Adriatic region are briefly discussed hereafter.

The first argument is based on recently published studies supporting the idea that Adriatic extreme events can only be captured with resolutions higher than those generally used in Regional Climate Models (RCMs). For the atmosphere, the capability of climate models to reproduce mesoscale wind phenomena has been tested and atmospheric resolution—via a better reproduction of the orography and consequently the enhancement of jet flows on finer grids, has been found to be one of the most important model characteristics known to impact wind speed driving extreme waves and storm surges in the Adriatic region (e.g. Belušić et al. 2017; Josipović et al. 2018). Similarly, for the ocean, extreme wave or flood hazard assessments have been found to only be achieved with models capable to reproduce the proper geomorphology (including complex coastline and bathymetry) of the Adriatic Sea coastal regions (e.g. Cavaleri et al. 2010, 2019; Torresan et al. 2019). Finally, based on the results of this study, the quantile–quantile plots (Fig. 17) displaying the performance of the unSWAN and SWAN 1-km models for an ensemble of 11 stations and 36 extreme events (presented in Sect. 2) illustrate the impact of the model resolution on the significant wave height and peak wave period. They show that for extreme conditions the unSWAN model provides better results than the SWAN 1-km model, particularly concerning the peak wave period, even though wave heights are slightly overestimated by the unSWAN model and perfectly represented by the SWAN 1-km between 1 and 3 m.

Fig. 17

Analysis of the sensitivity of the wave results to model resolution: with quantile–quantile distributions of the AdriSC SWAN 1-km and unSWAN results vs. measurements for the ensemble of 9 stations and 36 extreme events (left panels) and with spatial distributions of unSWAN results and differences between unSWAN and SWAN 1-km results for the maximum significant wave height and the maximum peak wave period during one extreme sirocco event (22 December 1979) and one extreme bora event (7 February 2012)

The second argument presented here is that the inclusion of the precise coastline and bathymetry (e.g. the precise geomorphology of the Venice Lagoon) as well as the numerous small islands (more than 1000 islands, isles, islets, rocks are identified along the Croatian coast) acting as barriers during extreme storm events, is crucial to properly simulate wave transformation and consequently storm surges in the northern Adriatic. This geomorphological impact is illustrated (Fig. 17) with spatial plots of maximum significant wave height and maximum peak wave period for the unSWAN model and for the difference between the unSWAN and SWAN 1-km models during the sirocco storm of the 22nd of December 1979 and the extreme bora event of the 7th of February 2012. The effect of the geomorphology and the islands located along the eastern Adriatic Sea on the wave propagation and transformation is particularly seen on the unSWAN results which show that both maximum significant wave height and maximum peak wave period are smaller along the Croatian coasts than in the rest of the domain. The difference plots highlight that, on the one hand, the overall increase in maximum significant wave heights in the unSWAN simulations (between 0.5 and 1.5 m compared to SWAN 1-km) mostly results from the increase in resolution (and thus in intensity) of the atmospheric wind forcing and, on the other hand, for the presented bora event, the proper representation of the islands in the unSWAN simulations dramatically influenced the reproduction of the maximum peak periods (up to 5 s difference with the SWAN 1-km model).

Finally, the use of unstructured meshes is not only useful for pure oceanographic purposes but also for the study of other processes driven by extreme events such as, for example, the motion of boulders during sirocco storms which requires to perfectly reproduce the wave transformation at a meter-scale resolution to derive the transport of these blocks weighting up to a ton (see Biolchi et al. 2019b).

Appendix 2

As no catalogue of extreme historical bora events exists in the Adriatic region, an in-depth bibliographical research—presented in Table 4 and including meteorological bulletins (in Croatian), local newspapers, photographs and videos, was thus undertaken to generate the ensemble of 22 historical bora storms used in this study.

Table 4 References used to select the date of the 22 representative extreme bora events of the ensemble