What spatial scales are believable for climate model projections of sea surface temperature?

Earth system models (ESMs) provide high resolution simulations of variables such as sea surface temperature (SST) that are often used in off-line biological impact models. Coral reef modellers have used such model outputs extensively to project both regional and global changes to coral growth and bleaching frequency. We assess model skill at capturing sub-regional climatologies and patterns of historical warming. This study uses an established wavelet-based spatial comparison technique to assess the skill of the coupled model intercomparison project phase 5 models to capture spatial SST patterns in coral regions. We show that models typically have medium to high skill at capturing climatological spatial patterns of SSTs within key coral regions, with model skill typically improving at larger spatial scales (a parts per thousand yen4A degrees). However models have much lower skill at modelling historical warming patters and are shown to often perform no better than chance at regional scales (e.g. Southeast Asian) and worse than chance at finer scales (< 8A degrees). Our findings suggest that output from current generation ESMs is not yet suitable for making sub-regional projections of change in coral bleaching frequency and other marine processes linked to SST warming.