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

observations. The range of values is between 0.25˚C to 0.4˚C, larger than GLORYS and closer to the observed range. Conversely, the CMCC-CMS model shows poor skills (RMSE of 0.09˚C and correlation -0.63). Trends The interannual variations and the long term evolution of the basin averaged SST anomalies showed by the two reference simulations were in good agreement with the observations both in the Red Sea and the Gulf of Aden (Fig 6). The observed linear trends (±> the standard error) for the period (1985–2005) were 0.021 ±>0.007˚C year-1 in the Red Sea and 0.002 ±>0.007˚C year-1 in the Gulf of Aden. As KAUST and GLORYS do not cover the same period, trends were not computed for these datasets. The GCMs showed trends ranging from 0˚- C�year-1 to 0.050˚C�year-1 in both regions, with an average value of 0.021˚C (S1 and S2 Tables). GCMs showing trends outside the range defined by the observed trend plus ± the uncertainty (square root of the i.e. quadratic sum, l264), were flagged as suspicious. Model selection The different diagnostics presented above had aimed at characterizing different aspects of the heat uptake, which is what determines the response of the Red Sea to the projected global warming. However, the quality of the GCMs was very heterogeneous, with some of them showing low skills in most of the diagnostics considered before. Therefore, prior to analyse their results for the climate projections, we have to select a subset of GCM models displaying reasonable skills in a selection of diagnostics. Unfortunately, there is no objective method that

ensures the “right” choice of diagnostics, as some models perform well in some of them while performing poorly in others. Therefore, we defined 4 different performance criteria based on a Fig 5. Interannual std of sea surface temperature from (a) observations, (b) the GLORYS reanalysis, (c) a GCM with good skills (IPSL-CM5a-LR;), (d) a GCM with low skills (CMCC-CMS;) and (e) the average of selected GCMs (see “Model selection” section). https://doi.org/10.1371/journal.pone.0255505.g005 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 10 / 26