Fig. 2: Temperature Variability Evaluation Matrices.

Temperature Variability Evaluation Matrix for different ocean (O1–O9) and land (L1–L24) regions for December, January, February (DJF) and June, July, August (JJA) months for the 11 single model initial-condition large ensembles (SMILES) for detrended (a) and non-detrended (b) temperature anomalies. Shading marks the fulfillment of Criteria 1, that the rank histogram of spatially averaged observations falls within the perfect-model rank range. Numbers mark the percentage of grid cells in the region that exhibit an unbiased representation of monthly surface air temperature (TAS) GISTEMPv4 observations over land and sea surface temperature (SST) ERSSTv5 observations over the oceans. An unbiased simulation at the grid cell level corresponds to observed monthly values exceeding the ensemble maxima or minima less than 8% of the months respectively, and occurring with the central 75th percentile ensemble bounds no more frequently than 80% of the months. A region and season are considered to be adequately simulated by a given model when it fulfils Criteria 1 (green and gray shading) and Criteria 2 (exhibiting at least 50% of the grid cells are unbiased). When both criteria are met the field is shaded in green, when only Criteria 1 is met the field is shaded in gray (see “Methods” for further details on the evaluation framework). The four bottom rows show the total number of regions that each model simulates adequately (TOTAL), as well as per season (T_Season) and across ocean (T_Ocean) and land (T_Land) regions, respectively. The last two columns show the number of models that adequately simulate a region per season. Regions for which less than 3 SMILEs offer adequate simulations are highlighted in red.