Identifying coral refugia from observationally weighted climate model ensembles

Reef-building corals face a variety of mounting anthropogenic stressors originating both at the global scale (increasing sea surface temperatures (SSTs), rising sea levels, and ocean acidification) and the local scale (such as destructive fishing, overfishing, sedimentation, invasive species, nutrient over-enrichment, and chemical pollutants). We propose a statistical downscaling method that can model multiple GCM outputs simultaneously together with observational data. The model is built hierarchically in a Bayesian framework and we incorporate the skills of GCMs from model weighting into the downscaling model, also producing uncertainty estimates across multiple GCMs. At each reef location, we use this model to estimate the year after which SST conditions annually exceed a severe bleaching threshold of 8°C-weeks, with the goal of guiding conservation management at local scales.