CHAOJIAO SUN: Deep learning projections of coral bleaching risk: will the 2024–25 Ningaloo marine heatwave become the new normal?
Coral bleaching poses an existential threat to reef ecosystems, yet global climate models operate at resolutions too coarse to resolve local-scale thermal dynamics. While dynamical downscaling provides higher-resolution solutions, it remains computationally prohibitive for multi-model, multi-scenario assessments. This study applies a deep learning downscaling framework to seven best-performing CMIP6 models for the Australian region, including ACCESS-ESM1.5 and ACCESS-CM2. By enhancing the utility of the ACCESS community’s Earth system outputs, this workflow provides a computationally efficient alternative to dynamical downscaling.
Raw Sea Surface Temperatures (SSTs) were bias-corrected using satellite observations to remove a systematic ~1°C warm bias, then downscaled to 10 km resolution using a Super-Resolution Deconvolution Network (SRDN) across three emission scenarios (SSP1-2.6, 2-4.5, and 3-7.0). Bleaching risk was quantified using Degree Heating Weeks (DHW).
By 2100, Ningaloo SSTs are projected to rise by 1.7°C, 2.5°C, and 3.7°C under the low, intermediate, and high scenarios, respectively. Annual maximum DHW is projected to more than double, from 6.1°C-weeks at 1.5°C global warming to 13.4°C-weeks at 2.0°C warming, consistently exceeding the severe bleaching threshold (8°C-weeks). These projections suggest that the unprecedented 2024–25 marine heatwave, which caused >75% coral loss in northern Ningaloo, may soon represent a routine occurrence, serving as a precursor to a new climatic normal. This research demonstrates the efficacy of deep learning for high-resolution coastal risk assessment, providing a scalable path for climate adaptation planning in sensitive World Heritage areas.
Keywords: coral bleaching, sea surface temperature, deep learning, downscaling, CMIP6, Ningaloo, Degree Heating Weeks, marine heatwave, climate projections