<p>Multi-model ensemble data (MMEs) are routinely used to improve reliability of future climate change projections. To derive MMEs, global circulation models (GCMs) are mostly selected for their ability to simulate mean climatology using temporal performance indices. However, choosing and ranking GCMs based on their explicit ability to reproduce spatial-temporal patterns of observed extreme climatology could be more intuitive for curtailing climate change threats. In this study, we ranked 11 and 13 GCMs included in the Coupled Model Intercomparison Project Phase 5 and 6, respectively, based on their performance in reproducing spatial-temporal patterns of extreme climatology in South Korea during 1975–2005. MMEs were derived by merging bias-corrected data from GCMs through simple mean and random forest (RF) algorithms. GCMs with finer spatial resolutions performed better in reproducing the spatial-temporal patterns of observed extreme climatology. RF-based ensembles derived by merging data from the three best-performing GCMs demonstrated superiority over the MMEs derived using simple mean, and these were used to project spatial-temporal trends of future extreme climatology. Projected decline rates of extreme cold events in South Korea were substantially faster than the intensification rates of extreme warming events. Rainfall-derived extreme indices projected intense/heavy monsoon rainfall events, followed by dry winters across the mid-latitude warmer inland and coastal regions of South Korea. The methodology presented in this study could be replicated to derive reliable MMEs for multi-sectoral climate change impact assessment and devising mitigation plans.</p>