A Machine Learning Approach to Rapidly Downscale Sea Surface Temperature Extremes and Heat Stress on the Great Barrier Reef

Abstract

Reef-scale climate projections, such as those generated by CMIP6, are critical for guiding the development of effective intervention strategies for mass coral bleaching events. We developed a machine learning (ML) model based on a super resolution deconvolutional neural network to rapidly downscale sea surface temperature (SST) on the Great Barrier Reef (GBR). When downscaling 80 km data to 10 km resolution, the ML model outperforms conventional interpolation methods by capturing the spatial variability of SST and extreme thermal events. We applied this model to independent datasets from both present-day and future climates, demonstrating its robustness. Additionally, we demonstrated the ML model's capability to reconstruct the spatial variability of degree heating weeks for coral bleaching risk analysis. With its ease of implementation and low computational cost, this ML model could be readily used or easily trained to rapidly downscale climate model outputs for coral reefs around the world.