The Adriatic Sea and Coast (AdriSC) kilometer-scale atmosphere-ocean climate model (Denamiel et al., 2019; Denamiel, 2021) has been specifically implemented to accurately represent the Adriatic coastal dynamics under both historical (1987-2017) and extreme warming (2070-2100; for a Representative Concentration Pathways RCP 8.5 greenhouse scenario) conditions. It is based on the Coupled Ocean–Atmosphere–Wave–Sediment Transport (COAWST) modelling system developed by Warner et al. (2010) which dynamically couples the Weather Research and Forecasting (WRF; Skamarock et al., 2005) atmospheric model and the Regional Ocean Modelling System (ROMS; Shchepetkin and McWilliams, 2009) ocean model. The AdriSC setup ( Table 1 ) consists in (1) two different nested grids at 15-km and 3-km resolution used in the WRF model and covering respectively the central Mediterranean area and the Adriatic-Ionian region and (2) two different nested grids at 3-km and 1-km resolution used in the ROMS model and covering respectively the Adriatic-Ionian region (similarly to the WRF 3-km grid) and the Adriatic Sea only. For the atmospheric grids, a 1:5 ratio is used to obtain slightly more accurate interpolation as the center of gravity of the outer grid cells is preserved within the inner grid (Chow et al., 2019). In addition, in climate modelling, due to multi-decadal runs, the 1:3 ratio is rarely used in order to preserve numerical resources; some experimental climate models even use 1:11 and 1:15 ratios (Zhou and Chow, 2014; Munoz-Esparza et al., 2017; Wiersema et al., 2018). In the vertical, all the AdriSC grids rely on terrain following coordinates: 58 levels refined in surface with a ratio of 3.62% of the height between the surface and 70 hPa (the top layer of the WRF model; Laprise, 1992) and 35 levels refined near both the sea surface and bottom floor for the ocean (Shchepetkin and McWilliams, 2009). More details on the set-up of the AdriSC modelling suite – which is installed and fully tested on the European Centre for Middle-range Weather Forecast (ECMWF) high-performance computing facilities – can be found in Denamiel et al. (2019); Denamiel et al. (2021b) and Pranić et al. (2021).

The AdriSC climate modelling strategy follows the Pseudo-Global Warming (PGW) method developed by Schär et al. (1996) in the atmosphere and recently extended to the ocean (Denamiel et al., 2020b). Consequently, climatological changes are derived based on the differences between future scenarios (here for the 2070-2100 period) with historical data (here for the 1987-2017 period). These changes are obtained using the LMDZ4-NEMOMED8 regional climate model, part of the Med-CORDEX ensemble (Hourdin et al., 2006; Beuvier et al., 2010), forced by the IPSL-CM5A-MR global climate model (simulations r1i1p1). These projections are imposed as forcing on top of the reanalysis products. Specifically, ERA-Interim (Balsamo et al., 2015) for the atmosphere, and the Mediterranean Forecasting System (MFS) MEDSEA v4.1 (Pinardi et al., 2003) for the ocean. The reanalysis products are the same ones used to produce the historical run (Denamiel et al., 2020a; Denamiel et al., 2020b).

The evaluation of the AdriSC climate model against an extensive dataset composed of in-situ observations and remote sensing products has shown that, for the 1987−2017 period, the skills of the newly developed coupled atmosphere-ocean kilometer-scale climate model outperform those of the RCMs implemented in the Mediterranean Sea (i.e., EURO- and Med-CORDEX) in both the atmosphere (Denamiel et al., 2021b) and the ocean (Pranić et al., 2021).

In this study, the far-future trends and variabilities over the entire Adriatic region as well as the number of extreme events are extracted from the AdriSC extreme warming run (2070−2100) and compare with the results of the historical run (1987-2017) already analyzed in Tojčić et al. (2023). They are derived for the surface WRF 3-km daily results – i.e., temperature at 2 m, rain, relative humidity at 2 m and wind speed at 10 m – in the atmosphere, and for the bottom, 100 m depth and surface ROMS 1-km daily results – temperature, salinity, and current speed – in the ocean.

Variances and trends are derived over the entire Adriatic region (WRF 3-km domain in the atmosphere and ROMS 1-km domain in the ocean). Trends are calculated over the 31-year period of both runs, with the Theil-Sen method (Mondal et al., 2012) and trend significances are calculated with the Mann-Kendall test (Mann, 1945; Kendall, 1975; Gilbert, 1987). Only trends with significance over 95% are taken into consideration and presented hereafter. Variances are calculated from the detrended daily data. The impact of climate change is calculated as percentages of trends/variances, computed as the ratio (multiplied by 100) of the difference between extreme and historical trends/variances and absolute value of historical trends/variances. The results are presented as spatial plots covering the whole studied areas. For convenience, the obtained original trends are converted to decadal trends, which are commonly used within the climate community.

The impact of climate change on the number of extreme events is derived for selected sub-domains. In the atmosphere, the variables are analyzed for the land and sea sub-domains in order to reflect the differences in heat capacity and heat transfer mechanisms. Indeed, land surfaces heat up and cool down more rapidly than sea masses, which results in distinct temperature gradients and air pressure variations. Consequently, local wind patterns, cloud formation, and precipitation distribution can significantly differ between land and sea areas. The ocean variables are analyzed for the deep and coastal sub-domains which cover the Southern Adriatic Pit ( Figure 1 ) and the deep areas surrounding it, and the rest of the Adriatic with relatively shallow depths, respectively. These sub-domains reflect the differences between the shallow and nearshore areas which may experience more significant temperature and salinity fluctuations in response to extreme weather events, river discharges, tides, etc. and the more stable deep areas less impacted by these processes. For each of the sub-domains, monthly means are derived from the extreme warming and historical AdriSC daily results. The monthly analysis is conducted through two distinct approaches applied to each studied variable. First, the difference between extreme warming and historical monthly climatologies is computed, serving as a basic indicator of the monthly changes in average conditions. Second, the 10^th and 90^th percentiles of the historical monthly data are determined, serving as minimum and maximum historical thresholds needed to calculate the number of days within a month below or above these thresholds (i.e., the number of extreme low or high events). To quantify the change in occurrences of extreme values under extreme warming conditions, compared to the historical ones, the number of days in the historical run with values below (or above) the minimum (or maximum) historical thresholds are subtracted from the number of days in the far-future climate. This represents approximately 3 days per month. The monthly analysis of the extreme events is not performed both at 100 m depth for the coastal sub-domain, as only a few model grid points in this domain fall below 100 m, and for the extreme low rain events as the minimum threshold is 0 mm/day at almost all model domain grid points.

Decadal temperature trends in the Adriatic region indicate that significant warming continues during the 2070-2100 period ( Figure 2 , upper panels). Temperature trends at 2 m are higher over sea (up to 0.4–0.5°C per decade) than over land (up to 0.3°C per decade) similarly to the trends found in the historical run (Tojčić et al., 2023). The percentage of temperature trend shows that far-future warming is generally expected to be less intense compared to the historical conditions. This is particularly visible over most of the Adriatic Sea (up to a 10% decrease) and the southern Italy and Dinarides (15% decrease). However, trends are projected to increase in the southern Pannonian plains (around 7%). Variance of temperature at 2 m above 35°C² is observed over land, with lower variability along the coast due to the influence of the sea and roughly two times less variability over the deep Adriatic Sea. Compared to the historical conditions, a general 5-10% increase in temperature variability is projected for this extreme warming scenario. Monthly analysis of the change in number of extreme air temperatures extreme events ( Figure 2 , lower panels) aligns with the spatial analysis, indicating more significant changes over sea, with over 20 more days per month over the historical maximum threshold and not a single day below the historical minimum threshold (i.e., about 3 less days per month than in the historical run), than over land. In particular, July-August exhibits nearly continuous extreme high temperatures over sea. Over land, around 10 days per month are projected to exceed the historical maximum threshold throughout the year, with over 20 days in July-August while no day below the minimum historical threshold is expected.

Relative humidity trends at 2 m ( Figure 3 , upper panels) are negative over sea (-0.6% to -0.4% per decade) and positive over land (0.3-0.5% per decade), mirroring present climate trends (Tojčić et al., 2023). These results probably originate from boundary condition forcing, as negative trends in relative humidity have been previously found in global analysis over most of the Mediterranean Sea (Vicente-Serrano et al., 2018). Negative trends are expected to intensify over the southern Adriatic Sea and most of the coastal areas but to decrease over the middle Adriatic Sea. Mountainous regions, like the Apennines and the Dinarides, exhibit patchy distributions, reflecting local effects of mountain ridges on humidity trends. Variability of relative humidity is expected to decrease in the far-future climate, ranging from 5% over sea to 15% in the northern Italy, Istria, and central Dalmatia coastal areas. Exceptions include mountainous regions like the Apennines, Sicily, and southeastern Dinarides. Monthly analysis for extreme relative humidity in the far-future climate ( Figure 3 , lower panels) reveals a decrease (less than a day per month) in dry days (below the minimum historical threshold) over land in September-December. This decrease is accompanied by an overall increase in mean relative humidity (up to 0.8%). Over sea, the decrease (less than one day per month) occurs in January-March. Moist days (above the maximum historical threshold) are expected to slightly decrease (less than a day per month) during the warm season over both land and sea sub-domains. The land-sea contrasts in terms of relative humidity are defined as positive climatology differences over land and negative ones over sea, except in May-August when both are negative.

Rain decadal trends in the far-future climate ( Figure 4 , upper panels) are positive over sea and the coastal regions, except in the Rijeka area where large negative trends are projected (below -0.2 mm/day per decade). Coastal regions along the eastern coast and the northern Adriatic present larger decadal rain trends (up to 0.2 mm/day per decade) than in the rest of the domain (around 0.1 mm/day per decade). South of the Adriatic, trends are lower and mostly decreasing. Rain variance percentages indicate an overall increase of more than 50% in variability. Monthly analysis of rain ( Figure 4 , lower panels) reveals an increase in days with extreme rainfall in the far-future climate, particularly in October-January. The number of days with extreme rainfall decreases (up to a day per month) during the warm season (May-September) over both land and sea sub-domains, accompanied by an increase in mean rain intensity over the land and sea sub-domains varying between 0.2 and 0.8 mm/day during the cold season (September-January).

For the wind speed at 10 m ( Figure 5 , upper panels), positive decadal trends are mostly observed over the sea and coastal regions (mostly between 0.05 and 0.1 m/s per decade), except along the bora wind jets, where trends are slightly negative (between -0.025 and 0 m/s per decade). Mountainous regions and the Pannonian plains show mostly negative trends (around -0.05 m/s per decade). Variability in wind speed is projected to decrease over most of the domain, reflecting greater persistence of wind regimes. Exceptions are mountainous regions, where variability is expected to increase by approximately 10%. Highest total variances, reaching up to 25 m² s-², in wind speed at 10 m are found along the Velebit mountain, known for its strong bora wind events. Monthly wind speed analysis ( Figure 5 , lower panels) indicates a decrease in strong wind events throughout the year over the land domain, except in October-November. Over sea, a decrease is noted between January and March. Winter sees the most significant decrease in mean wind speed over both land and sea domains (up to -0.3 m/s), while July-August experiences a small increase over sea. The decrease of wind speed during winter, particularly in the northern part of the domain, might suggest a decrease in bora winds and turbulent heat flux, which might impact the northern Adriatic dense water formation (see Section 4 for detailed explanation).

Sea surface temperature trends ( Figure 6 ) predominantly range from 0.25 to 0.35°C per decade, with slightly higher values in the southern than in the northern Adriatic and maximum values at the perimeter of the Southern Adriatic Pit (SAP). Trend percentages ( Figure 6 ) indicate a decrease in sea surface temperature trends in the far-future climate compare to the historical results, mostly around 30%, with higher values of approximately 50% in the central Adriatic and along the western coastal current but lower values below 10% in the SAP. Sea temperature trends at 100 m depth ( Figure 7 , upper panels) show an increase of approximately 0.4°C per decade. The Otranto Strait experiences the highest warming rates (around 0.5°C per decade), while the SAP shows the lower values (about 0.35°C per decade). Trend percentages indicate a decrease in temperature trends of 25-40% in the central Adriatic and an increase of 25-50% in the SAP and the Otranto Strait compare to the historical results. Decadal trends of sea bottom temperature ( Figure 8 , upper panels) demonstrate contrasting behavior between the SAP and the rest of the Adriatic Sea. The SAP exhibits negative trends (0.1-0.2°C per decade), while positive trends (up to 0.5°C per decade) are observed elsewhere, with the highest values along the eastern coast and the Dalmatian islands. Trend percentages indicate a general decrease in future sea bottom temperature trends, with the lowest values in the central Adriatic and along the Dalmatian coast and islands, dropping below 20%. Higher percentages are found in the SAP and the Otranto Strait, with trends up to 200% lower than for the historical results. Far-future temperature variability is expected to increase compare to the historical results, by 10-20% at the surface and up to 100% at 100 m in the SAP and the Otranto Strait. Monthly analysis reveals significant changes in sea temperature extremes in the far-future climate. Temperatures below the minimum historical threshold in the historical run no longer occur but temperatures above the maximum historical threshold persist in all months, except May-June when 5-8 days do not exceed this threshold. Differences in monthly climatologies show a sea surface temperature increase in the coastal Adriatic regions ranging from 3.4-3.5°C in winter to 4.1-4.4°C in summer. At 100 m, the increase is of 3.4-3.6°C, more pronounced in winter and less in spring and summer due to seasonal conditions. At the bottom, temperature differences also vary, with coastal areas experiencing similar values to those at 100 m, but significantly lower values in the deep sub-domain, ranging from 0.9 to 1°C.

Surface salinity trends ( Figure 9 upper panels) generally vary around 0.1 per decade, with slightly lower values in the Kvarner Bay and Dalmatian Island regions (0.05 per decade). Higher salinity trends are observed off the Po River delta (exceeding 0.15 per decade) and in the Otranto Strait. Compared to the historical trends, salinity is expected to increase by approximately 25% in the SAP and the eastern side of the Otranto Strait. Lower salinity trend differences are projected for the rest of the Adriatic, particularly along the Po River plume, the Kvarner Bay coastal area, and the Neretva River delta. At a depth of 100 m ( Figure 10 ), salinity trends range between 0.05 and 0.1 per decade while, compare to the historical trends, far-future trends are expected to decrease by approximately 20% in the western Adriatic region but to increase in the eastern part, especially in the SAP. Variability at 100 m is generally low but is expected to be much larger than in the present climate. Bottom salinity trends ( Figure 11 upper panels) are positive, with lower values in deeper regions and higher values in shallower areas, such as the northern Adriatic shelf. Salinity trends decrease along the Po River plume, in the northern Adriatic, and the Kvarner Bay, while they increase in the Southern Adriatic Pit and the Otranto Strait. Bottom salinity variability is projected to decrease in the Southern Adriatic Pit but increase in the rest of the Adriatic. At the surface ( Figure 9 ), differences in the number of days of extreme high salinity events reveal positive changes during summer (up to 3 and 2 days more in the coastal and deep domains, respectively), but negative changes, associated with an average decrease in salinity of up to 0.2, otherwise. At 100 m depth, all differences are negative throughout the year with at least 14 days more of extreme low salinity events than under the historical conditions associated with a decrease in salinity up to 0.15. This clearly indicates lower salinity in the far-future climate, especially during the winter months. At the bottom, differences are negative in the coastal sub-domain leading to at least 4 days more of extreme low salinity events, while in the deep domain, they are positive leading to at least 18 days more of extreme high salinity events. A detailed discussion about the potential impact of climate warming on the salinity budget is presented in Section 4.

Far-future current speeds in the Adriatic Sea ( Figures 12 – 14 ) show positive trends across all depths, like it has been documented for the present climate (Tojčić et al., 2023). At the surface ( Figure 12 ), trends range from 0.005 to 0.01 m/s per decade in most areas, except for the SAP, where they reach up to 0.02 m/s per decade in the inner part of the cyclonic gyre and around -0.005 m/s per decade in its outer parts. Compared to historical trends, positive changes are observed in the central SAP, transversely over the northern Adriatic, and along the eastern and western coastal regions. At a depth of 100 m ( Figure 13 ), current speed trends are similar to surface trends, with positive values up to 0.005 m/s per decade, primarily concentrated in the center of the SAP where both trends and variability increase by more than 100%. At the bottom ( Figure 14 ), current speed trends remain generally low, below 0.002 m/s per decade, with the lowest values in the SAP. Compared to the historical trends, a strong increase in trends (more than 100%) associated with a strong decrease in variability (up to 100%) is expected in the SAP and the Otranto Strait. Monthly analysis of surface current speeds indicates a significant increase in the deep domain during wintertime, while summer months show lower increases. Consequently, the number of days with extreme high surface current speeds will double or triple in the deep domain in the far-future climate, while weak surface current events will be decrease by half. At 100 m depth, mean current speeds will increase, particularly in January-May, with almost half of the month characterized by extreme high current speeds. At the bottom, the deep domain will experience a weakening of the currents, particularly in summer when up to 11 days more of extreme weak events per month are expected to occur compared to the historical conditions. In the coastal domain, the changes in current speed are mostly negligible.

The presented AdriSC results reveal several changes that can impact the atmosphere-ocean dynamics across the Adriatic region. These changes, summarized in Table 2 , are presented below for four different categories: (1) heatwaves, extreme rainfalls, and droughts, (2) dense water formation, (3) salinity budget and (4) Southern Adriatic dynamics.

Two different ensembles of regional climate models (RCMs) projecting climate change under RCP8.5 conditions are available in the Mediterranean region. The EURO-CORDEX ensemble (Coppola et al., 2021) at 0.11° resolution with 14 different RCMs driven by 8 CMIP5 global climate models (i.e., 68 members in total), is mostly composed of atmospheric models (few coupled with ocean models). The Med-CORDEX ensemble (Ruti et al., 2016) mostly composed of atmosphere-land-ocean models is formed of 17 Atmosphere RCMs at 10-50 km resolution, 8 Ocean RCMs at 7-8 km resolution, 4 Land-Surface RCMs and 14 Fully coupled RCMs at about 10 km resolution, driven by at least one of the 8 different CMIP5 global climate models use in EURO-CORDEX. As trends and variability are generally well reproduced by RCMs, these ensembles can thus be used to perform a comparative evaluation of the far-future AdriSC projections (see Table 2 ).

Ivušić et al. (2021) analyzed the EURO-CORDEX ensemble precipitations over the Adriatic region for the far-future period (2071-2100) with respect to the historical period (1971-2000). They found a considerable reduction of the total precipitations during the summer months associated with a decrease of the number of rainy days over the entire region and strong south-north gradients particularly during the winter months when the precipitation intensity could increase. Baronetti et al. (2022) analyzed both EURO- and Med- CORDEX simulations in northern Italy for the same periods and also found a north-south spatial gradient as well as an intensification of the droughts. As in EURO- and Med- CORDEX simulations, under the far-future AdriSC projections, the number of days with extreme precipitations is also expected to decrease during summer and to increase during winter over both the land and sea domains. In terms of south-north gradients, the most intense positive precipitation trends are mostly located over the Adriatic Sea and below 44°N of latitude, while the negative trends cover the Pannonian plain mostly above 44°N of latitude. However, following the trends and variability analyses, the land-sea contrasts also highlighted in Coppola et al. (2021) are much more pronounced than the south-north gradients in the AdriSC results.

Belušić Vozila et al. (2019) used two members of the EURO-CORDEX ensemble to analyze the far-future near-surface winds over the Adriatic region. They found a reduction in the number of bora events, an increase in the number of sirocco events in the northern Adriatic during the winter season, and a decrease of the intensity of both bora and sirocco events, except in the northern Adriatic for the bora events. The AdriSC far-future results also project a decrease in the intensity of the wind speeds over the entire Adriatic region particularly during winter time as well as a decrease in the number of extreme events above the maximum historical threshold over the sea sub-domain particularly during January-April which is the period during which strong bora events occur. The increase in sirocco events cannot be seen in the presented AdriSC results which average the wind speed conditions over the entire sea sub-domain.

Parras-Berrocal et al. (2023) analyzed the REMO-OASIS-MPIOM model results (Sein et al., 2015) of the Med-CORDEX ensemble and found that the projected dense water formation could be reduced by 75 % in the Adriatic Sea (84 % in the Aegean Sea, 83 % in the Levantine Sea) by the end of the century due to hydrographic changes in surface and intermediate water that strengthen the vertical stratification, hampering vertical mixing and thus convection. As documented by Soto-Navarro et al. (2020), the changes will manifest in an increase of temperature over the whole water column, more at the surface and less in deep layers, while most Med-CORDEX models are projecting an increase in salinity of the whole Adriatic for the RCP8.5 scenario. The AdriSC results are mostly aligned with these findings. First, the northern Adriatic dense water formation is expected to be reduced due to local changes (increased sea surface temperatures, decreased sea surface salinity, decreased wind speeds). Second, the changes in properties of the intermediate (decreased salinity and increased temperatures) and bottom (increased salinity and increased temperatures by 2.5°C less than in the intermediate layer) layers over the SAP seen by the AdriSC model also confirm a strengthening of the vertical stratification potentially linked to the Ionian-Adriatic exchanges.

Finally, the limitations of the AdriSC far-future projections are twofold. First, the PGW method uses the same synoptic forcing in both historical and extreme warming runs which implies that potential changes in intra- and interannual variability might be missed. Second, as the results of only one RCM are used in the AdriSC PGW approach, the climate uncertainty, derived by running ensembles of simulations forced by multiple global climate models under multiple warming scenarios (Semenov and Stratonovitch, 2010), is ignored.