Assessment of Red Sea temperatures in CMIP5 models for present and future climate

(including the Gulf of Suez and the Gulf of Aqaba) and part of the Gulf of Aden (area enclosed by 10–30˚N and 30–50˚E) but does not cover the Outer Region. The model is forced by ECMWF (European Centre for Medium-Range Weather Forecasts) atmospheric fields [32]. It provides monthly mean temperatures with a spatial resolution of 0.1˚ and 50 vertical z-levels for the period (2001–2017). This model is developed as an updated version of the one imple- mented by [21] to investigate the general and overturning circulation in the Red Sea. The model outputs were validated against different observational datasets including temperature and have been used in various Red Sea studies. To name but a few, it has been used to analyse the eddy field [33], the phytoplankton phenology in the South of the Red Sea [34] and in the North [35] or the general connectivity [36]. In order to complement the KAUST model and to also cover the Gulf of Aden, the GLORYS.S2V4 global reanalysis was also analyzed [17,20]. This product was generated by NEMOv3.1 ocean model with a horizontal resolution of 0.25˚ and 75 vertical z-levels. It is forced by ERA-Interim atmospheric fields [32], for the period 1993 to 2015. GLORYS assimi- lates along-track satellite observations of sea level anomaly, sea ice concentration and SST, and in situ profiles of temperature and salinity from the CORA data base [37]. The data is available at (http://marine.copernicus.eu/services-portfolio/access-to-products/?option=com_ csw&view=details&product_id=GLOBAL_REANALYSIS_PHY_001_025). Indicators for model skills

A set of diagnostics to rank the performance of the models were computed from the model outputs and the observational datasets for the period 1985–2005 for the Red Sea and the Gulf of Aden (Outer Region) separately (Fig 1, S1 and S2 Tables). Those diagnostics have been cho- sen to assess the skills of the models in reproducing different aspects of the heat content vari- ability, which is what will determine the evolution of Red Sea temperatures under the projected global warming. The detailed formulation of each diagnostic is included in the Sup- plementary Information. All the diagnostics have been performed into the native model grid. However, for ensemble averaging or mapping all model fields have been bilinearly reinterpo- lated into a common grid. For a clearer visualization, we have chosen the KAUST model grid as the reference grid, as it is the finest one. In order to fill the coastline gaps in GCMs (i.e. due to their coarse resolution), we have extrapolated the values from the closest GCM grid point. Area and volume. The spatial resolution of GCMs ranges from 0.2˚ to 2˚, so their repre- sentation of the Red Sea bathymetry may not be satisfactory. To quantify that we computed the area and volume of the Red Sea as represented by each model grid, and compare to those derived from the GEBCO database (General Bathymetric Chart of the Oceans), available at a high spatial resolution of 1/60˚ (https://www.gebco.net/data_and_products/gridded_ bathymetry_data). Models showing a difference in the area or volume larger than +25%

respect to the observed values, area (volume) for the Red Sea (i.e., ~4.57 10^5 km2 (2.18 ^5 km3)) and for the Outer Region (i.e., ~1.18 ^6 km2 (3.23 ^6 km2)), were discarded (see Sec- tion 3). Seasonal cycle of the averaged vertical profile. A key element for a robust modelling of future Red Sea thermal regimes is the characterization of heat uptake at the sea surface and the heat transfer to deeper layers. In order to characterize the skills of the models on reproducing this, we computed the averaged vertical profile of seasonal temperature anomalies with respect to the annual average for the upper 400 meters and compared modelled and observed profiles. Seasonal comparisons of the averaged vertical profiles were based in terms of correlation coef- ficient and the root mean square error (henceforth RMSE), thus assessing the shape of the pro- file and the magnitude of the anomalies, respectively. The four values obtained (i.e., one for PLOS ONE Assessment of Red Sea temperatures in CMIP5 models for present and future climate PLOS ONE | https://doi.org/10.1371/journal.pone.0255505 July 30, 2021 5 / 26