Deep learning to downscale future climate projections to assess future coral bleaching risks for the Ningaloo Reef

The unprecedented 2025 coral bleaching event in the Ningaloo Coast World Heritage Area highlights that even climate refugia are vulnerable to severe thermal stress. Here we use a deep learning approach to downscale sea surface temperature (SST) from five select CMIP6 models to a 10 km resolution. The quantile delta mapping method is used to correct ~1°C warm bias in SST from climate models determined by satellite observations. We used the NOAA SST satellite observations as the training dataset to resolve coastal regions in the Ningaloo and the Exmouth Gulf. We compare the results with those obtained using the SST training data from an eddy resolving ocean model. Our projections show that SST in the Ningaloo will warm by about 0.5°C at a 1.5°C global warming level, increasing to over 1.5°C at a 3°C level. This projected warming leads to a substantial increase in Degree Heating Weeks (DHWs), suggesting that the coral bleaching event of 2025 will likely become more common in the future.