Wintertime cooling on the shelf and deep convection process, common for the Adriatic Sea, are recognized as major drivers of the Adriatic-Ionian thermohaline circulation (Orlić et al., 2006). To observe, simulate and understand climate-scale variability of the thermohaline circulation and its drivers is still very challenging. Because of geographical position of the Adriatic Sea and the surrounding topography, the bursts of cold and dry Bora wind blowing over the basin (Grisogono and Belušić, 2009) are responsible for the Dense Water Formation (DWF) processes, in particular for: -Very cold and dense Northern Adriatic Dense Water (NAdDW) generated at shallow northern Adriatic shelf. After its formation, NAdDW is advected towards southeast inform of a bottom density current that travels along the western shelf and fills the deepest parts of the Adriatic (Jabuka Pit and South Adriatic Pit). - Adriatic Deep Water (ADW) that is formed through deep convection inside the South Adriatic cyclonic gyre. Joined with NAdDW, through Otranto Strait, ADW flows towards the Ionian Sea, where it fills the deepest layers of the Eastern Mediterranean (Zore-Armanda, 1963; Schlitzer et al., 1991; Artegiani et al., 1997a; Vilibić et al., 2004). Adriatic deep water masses are important for several reasons. As mentioned, they contribute to the Eastern Mediterranean circulation sustainability (Roether and Schlitzer, 1991; Manca et al., 2002) and refresh the deep layers of that area by bringing oxygenized waters (Malanotte-Rizzolii Robinson, 1988). They drive the thermohaline circulation of the Adriatic-Ionian basin (Orlić et al., 2006; Vilibić et al., 2013), and influence observed decadal oscillation of the northern Ionian Sea circulation (Borzelli et al., 2009), which in turn affects the dynamics of entire middle and eastern Mediterranean, especially of the Adriatic Sea (Gačić et al., 2010). The feedback mechanism between the Adriatic DWF and the circulation patterns of northern Ionian is called The Adriatic-Ionian Bimodal Oscillating System (BiOS), and is recognized as the main feature that drives the decadal variability of the Adriatic thermohaline properties (Mihanović et al., 2015). In this PhD research regional climate modelling approach is used to study thermohaline properties and variability of the Adriatic Sea. The research includes qualitative and quantitative analysis of the Adriatic thermohaline circulation and its variability, focusing on the processes of the DWF and the BiOS. Past experiences in modelling studies suggest the important issues to be implemented in order to properly reproduce the Adriatic-Ionian ocean processes, such as the DWF and the BiOS: -proper introduction of topography and bathymetry (Hendershot and Rizzoli, 1976), -high spatial and temporal resolution of the ocean (Pinardi et al., 1996) and atmospheric (Bergamasco et al., 1999; Beg-Paklar et al., 2001) components of a model, -proper introduction of buoyancy fluxes (Vested et al., 1998; Raicich et al., 2013) and river discharges (Janeković et al., 2014), and -appropriate boundary conditions (Mantziafou and Lascaratos 2004, 2008; Oddo and Guarnieri, 2011). However, all these researches were conducted with short-term numerical simulations, for which their results were limited only to the evaluation of the Adriatic-Ionian ocean processes on shorter timescales (up to 10 years). In order to quantify long-term changes of oceanographic properties related to the Adriatic-Ionian thermohaline circulation, and to test the performance of climate models with different setup in the basin (following previous suggestions), seven different regional hindcast simulations based on various configurations of NEMO–Mediterranean versions covering ERA-Interim period (1980-2012) were tested. This work is also testing the performance of coupled vs. non-coupled modelling approach. The chosen simulations of regional climate models are produced within the framework of the Med-CORDEX initiative (www.medcordex.eu), and are differing in their vertical and horizontal resolution, freshwater load, surface heat fluxes, air-sea interaction, and inclusion or not of the aerosol trend. The objective of this research is to evaluate the reliability of regional climate models for the Mediterranean region to reproduce the Adriatic-Ionian ocean dynamics, through their validation on in situ observations, and through detailed assessment of both Adriatic DWF processes and the BiOS, as well as their multi-decadal variability. The goal is to find the optimal model which will be the most reliable for quantification of Adriatic-Ionian thermohaline circulation in the future. Performance of all seven simulations is evaluated on the long-term in situ data collected over three northern Adriatic transects, along the Palagruža Sill transect, at Jabuka Pit and at South Adriatic Pit, as well as on the altimetry satellite observations. Detailed description of the Adriatic-Ionian thermohaline properties and dynamics is provided in Section 1. Section 2 covers the description of all used simulations, in situ and altimetry obtained data, and details of the performed analysis. A performance analysis of the NEMOMED8 ocean regional circulation model is given in Section 3. The analysis was undertaken for the Adriatic Sea during the period of 1961–2012, focusing on two mechanisms: the DWF and the BiOS, which drive interannual and decadal variability in the basin. The model was verified on sea surface temperature, sea surface height and long-termhydrographic in situ observations from several key areas. NEMOMED8 simulation qualitatively reproduces basin-scale processes, in particular: -thermohaline-driven cyclonic circulation and freshwater surface outflow along the western Adriatic coast, -dense water dynamics, and -the inflow of Ionian and Levantine waters to the Adriatic. However, positive temperature and salinity biases are reported; the latter particularly large along the eastern part of the basin, presumably because of the inappropriate introduction of eastern Adriatic rivers into the model. The highest positive temperature biases in the vertical direction were found in dense water collectors in the Adriatic, i.e. Jabuka Pit and South Adriatic Pit, indicating either inappropriate quantification of DWF processes or temperature overestimation of modelled dense water, especially of NAdDW. Moreover, much reduced model domain compared to the real bathymetry, which does not include entire coastal area of the northern Adriatic where the largest heat losses are found, i.e. the part of the Kvarner Bay (Janeković et al., 2014), also led to an underestimation of the cumulative surface heat losses during wintertime cold outbreaks, and therefore to the under representation of the NAdDW formation. As a consequence, the simulated NAdDW is of lower volume when it comes to the deep layers of South Adriatic Pit. Thus, an over estimation in vertical mixing of water column during deep convection process has been documented by the NEMOMED8 model. The DWF rates are qualitatively well reproduced by the model, being larger when preconditioned by higher basin-wide salinities. The decadal variability in the thermohaline properties is reproduced better than interannual variability, which is considerably underestimated. However, a key process that drives the Adriatic decadal variability, the BiOS, is not properly reproduced, in particular current reversals and outreach of the Northern Ionian gyre. The only appearance of anticyclonic circulation simulated by the NEMOMED8in the northern Ionian Sea was found only during the Eastern Mediterranean Transient. Next, performance analysis of seven regional ocean configurations based on NEMO has been carried out for the Adriatic Sea over a common period (1980-2012). The goal of the study was to test the model performances with different settings, particulary when it comes to reproduction of the Adriatic-Ionian thermohaline properties and variability. The analysis is given in Section 4. Simulations differ in resolution, model physics, atmospheric forcing (forced vs. coupled models) and river discharges imposed within the Adriatic Sea. Models have been evaluated on the long-term temperature and salinity measurements in all of the Adriatic sub-basins, in particular within dense water collectors (Jabuka Pit and South Adriatic Pit) and dense water formation sites (northern and southern Adriatic). Adriatic-wide salinity values are mostly linked to the proper introduction of the overall water budget, rather than to the local river forcing. Forced models mostly overestimate temperature and salinity values. On average, coupled models better reproduce the thermohaline properties and processes, in particular the BiOS reversals and its decadal variability. Wintertime heat losses are playing major role in defining the ADW transport rates in coupled models, while preconditioning in salinity is the most important factor in forced models. Further on, increase of resolution of the atmospheric forcing results in more realistic ocean dynamics, including the DWF in the complex coastal northern Adriatic. However, all models have large positive temperature biases at the dense water collector sites, indicating overall underrepresentation of the Adriatic DWF. Consequently, mixed-layer depth in the southern Adriatic is overestimated, reaching the bottom during some years and in some models. Ocean model resolution and river forcing seem to play a second-order role in defining the overall Adriatic-Ionian thermohaline properties, while inclusion of aerosol trend only slightly modified the reproduction of the BiOS. Lastly, a preliminary analysis of future projections which contain the very first attempt to quantify the processes of the DWF and the BiOSin the future climate, is described in Section 5. This also applies to the evolution of surface temperatures and salinities. The NEMOMED8 simulations were forced by three future scenarios of Representative Concentration Pathways (RCP, IPCC, 2013): RCP2.6, RCP4.5 and RCP8.5. The analysis is performed for the three future periods: near (2011-2040), middle (2041-2070), and far future (2071-2100), with respect to the last 30 years of the referent period simulated in historical simulation (1976 -2005). No significant trends in all analyzed parameters were found in control simulation covering the period 1950-2100. Towards the end of 21st century, under all three scenarios, gradual increase in basin-wide sea surface temperature (SST) was found, of which RCP8.5 is resulting with highest SST at the end of the 21st century, about 2.7 oC on average. All scenarios result also in an increased sea surface salinity (SSS), more pronounced at the coastal parts of the Adriatic, with highest increase found in RCP8.5. All three scenarios show a decrease in maximum mixed layer depth (MLD), suggesting a weakening of deep convection processes by the end of the 21stcentury. As for the analysis of the BiOS related processes, all scenarios are projecting an increase in intensity of the positive BiOS index (anticyclonic circulation in northern Ionian Sea), which favors the advection of less saline water masses coming from the Western Mediterranean to the Adriatic. The strengthening of the both positive and negative BiOS regimes is only projected with RCP8.5 scenario, more pronounced in the anticyclonic phase.