Modelling present and future climate in the Mediterranean Sea: a focus on sea-level change

Abstract

We present results of three simulations of the Mediterranean Sea climate: a hindcast, a historical run, and a RCP8.5 scenario simulation reaching the year 2100. The simulations are performed with MED16, a new, tide-including implementation of the MITgcm model, which covers the Mediterranean—Black Sea system with a resolution of 1/16°, further increased at the Gibraltar and Turkish Straits. Validation of the hindcast simulation against observations and numerical reanalyses has given excellent results, proving that the model is also capable of reproducing near-shore sea level variations. Moreover, the spatial structure of the elevation field compares well with altimetric observations, especially in the Western basin, due to the use of improved sea level information at the Atlantic lateral boundary and to the adequate treatment of the complex, hydraulically driven dynamics across the Gibraltar Strait. Under the RCP8.5 future scenario, the temperature is projected to generally increase while the surface salinity decreases in the portion of the Mediterranean affected by the penetration of the Atlantic stream, and increases elsewhere. The warming of sea waters results in the partial inhibition of deep-water formation. The scenario simulation allows for a detailed characterization of the regional patterns of future sea level, arising from ocean dynamics, and indicates a relative sinking of the Mediterranean with respect to the Atlantic more pronounced than the current one.

Appendix

The biases between the model projections and the reference reanalyses reported in Table 2 are comparable to the accuracy of the reference data, which is an issue in itself, and is reported in Table 7, having been estimated as the difference with respect to the alternative MEDHYMAP data. In addition, model precision is also roughly evaluated in terms of the Med-Cordex ensemble spread, by assuming that the ensemble mean is the best estimator for the expected value of each variable, and that different model realizations represent its alternative measurements. The values for historical experiments reported in Table 7 (first row) are inferred from Figs. 4 and 7 in Soto-Navarro et al. (2020), while for hindcast runs the only reference reported in literature is, to our knowledge, the basin-wide, total-depth temperature estimate illustrated in Figure 1 in Harzallah et al. (2018), which ranges from ~ 0.3 to ~ 0.4 °C for the considered period. The same figure also highlights the extent of possible differences between alternative observational products—in this case, RIXEN from Rixen et al. (2005) and EN₄ from Good et al. (2013)—which for the period of interest reach up to ~ 0.1 °C, compared to an associated uncertainty that can be twice as large.

Table 7 Model precision and observation accuracy

The distance between the reanalysis and their partly constraining reference observations over the period 1987–2021 is reported in Table 8, as derived from the Quality Information Document for the Copernicus Reanalysis (https://catalogue.marine.copernicus.eu/documents/QUID/CMEMS-MED-QUID-006-004.pdf). Results for the deepest layer suffer from scarcity of data and more pronounced model error. For the sake of completeness, we recall here that MEDHYMAP data are derived via a variational interpolation of observation, as the RIXEN data are, while the EN₄ dataset is the result of Optimal Interpolation. The Copernicus Reanalysis is computed by assimilating observations in the operative NEMO model runs via a 3DVar scheme, and therefore represents their dynamical interpolation.

Table 8 Distance between the reanalysis and their partly constraining reference observations