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

largest values in the southern Red Sea and the lower values in the Outer Region. In particular AAMI would reach about 150 ± 35˚C�day under scenario RCP4.5, with maximum values in the southern Red Sea (180 ± 55˚C�day), decreasing to 50 ± 30˚C�day approximately in the Outer Region (S9H and S9I Fig). Under scenario RCP8.5 the projected AAMI is 300– 400 ± 100˚C�day in the northern Red Sea and around 250 ± 120˚C�day in the Outer Region (Fig 10H and 10I). Discussion GCMs are not designed to simulate regional processes, mainly because of their coarse resolu- tion and the lack of specific tuning of model parameters, which is the main reason why GCMs are often discarded for regional analyses in semi enclosed seas [39]. Previous studies compar- ing regional and global models with observations suggest that regional simulations are required to properly simulate key processes in semi enclosed seas [16,40]. That would be the case for those analyses focused on the circulation and the 3D heat and salt distribution, which require a minimum spatial resolution to be properly solved. However, [18] showed that a rela- tively simple process like the air-sea interaction, can be reasonably reproduced by GCMs. Therefore, ocean variability which would only depend on the air-sea interaction could poten- tially be faithfully represented by GCMs. In this sense, [18] reported that projected changes of surface temperature in the Mediterranean for the 21st century were similar using an ensemble of GCMs or regional climate models. In conclusion, the suitability of GCMs for a particular

study is case dependent. In order to further investigate the driving mechanism for the Red Sea heat content variability we compare observed heat content in the upper 100 m in the Red Sea from the TEMPERSEA product [17] with air temperatures at 1000 mbars obtained from the JRA55 reanalysis [16], (https://rda.ucar.edu/datasets/ds628.1/?hash=access). The comparison of the yearly values for the period 1958–2013 (Fig 11) shows that interannual variations in the Red Sea heat content are highly correlated to variations in the air temperature (r = 0.68). It has to be noted that both datasets are totally independent as one is constructed from ocean obser- vations and the other using atmospheric data. This result reinforces the idea that heat content variations could be reproduced, at least as a first approximation, by models correctly reproduc- ing air-sea interactions. In order to assess if GCMs are able to reproduce the heat content variability in the Red Sea, here we have assessed the skills of 43 GCMs in terms of different diagnostics related to heat uptake. Our results are very heterogeneous with some of the models that did not even repre- sent the Red Sea in their ocean mask, and others that showed very poor results in all the diag- nostics. However, we have also found that some GCMs were able to reproduce the present variability with an accuracy comparable to the reference regional simulations (KAUST and GLORYS), especially inside the Red Sea. Those models are able to approximate reasonably

well the seasonal and interannual variability of the temperature profile in the upper 300 m, suggesting that ocean heat uptake is properly modelled. Also, they are able to represent the spatial variability with the north to south gradients in the Red Sea and the distinctive behav- iour of the Outer Region. We found the skill of GCMs to be typically better inside the Red Sea than in the Outer zone. This is consistent with a previous study [17], that has shown that the heat content evolution in the Red Sea basin is mainly driven by the air-sea interactions, while the Outer Region is also affected by the advection of waters from the open Indian ocean. This suggests that the mecha- nisms underpinning the basin temperature evolution of the Red Sea are relatively simpler, and thereby easier to be captured by GCMs. In the Outer Region, GCMs should be able to properly model not only the air-sea interactions but also the temperature evolution and the circulation 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 18 / 26