Performance of pattern-scaled climate projections under high-end warming, part I: surface air temperature over land

Pattern scaling is widely used to create climate change projections to investigate future impacts. We consider the performance of pattern scaling for emulating the HadGEM2-ES general circulation model (GCM) paying particular attention to ‘high-end’ warming scenarios and to different choices of GCM simulations used to diagnose the climate change patterns. We demonstrate that evaluating pattern-scaling projections by comparing them with GCM simulations containing unforced variability gives a significantly less favourable view of the actual performance of pattern scaling. Using a four-member initial-condition ensemble of HadGEM2-ES simulations, we infer that the root-mean-squared errors of pattern-scaled monthly temperature changes over land are less than 0.25°C for global warming up to ~3.5°C. Some regional errors are larger than this and, for this GCM, there is a tendency for pattern scaling to underestimate warming over land. For warming above 3.5°C, the pattern-scaled projection errors grow but remain small relative to the climate change signal.

We investigate whether patterns diagnosed by pooling GCM experiments from several scenarios are suitable for emulating the GCM under a high-end warming scenario. For global warming up to 3.5°C, pattern scaling using this pooled pattern closely emulates GCM simulations. For warming beyond 3.5°C, pattern-scaling performance is notably improved by using patterns diagnosed only from the high-forcing RCP8.5 scenario. Assessments of climate change impacts under ‘high-end’ warming using pattern-scaling projections could be improved by using change patterns diagnosed from pooled scenarios for projections up to 3.5°C above pre-industrial levels and patterns diagnosed from only strong forcing simulations for projecting beyond that. Similar findings are obtained for five other GCMs.