Deep Learning-Based Statistical Downscaling of Sea Surface Temperature Using a Residual Corrective Neural Network

The large-scale oceanic and atmospheric forecasts provided by global climate models typically lack sufficient resolution to accurately capture the response of the coastal ocean. Dynamical downscaling is computationally prohibitive, especially when applied to extensive coastlines, to many predictive ensembles, or long time periods. Therefore, this work presents a statistical downscaling of sea surface temperature (SST) from the seasonal coupled ocean-atmosphere forecast system (ACCESS-S2) using machine learning techniques.

This study proposes a novel two-stage deep learning framework that combines a U-Net convolutional neural network for SST prediction with a Residual Corrective Neural Network (RCNN) for iterative refinement toward high-resolution outputs. The target SST fields are derived from the Regional Ocean Modeling System (ROMS).
The RCNN progressively refines its predictions by incorporating dynamically scaled residuals at each step, enabling accurate capture of both broad patterns and fine-grained features such as eddies and fronts.
We also introduce a custom loss-assisted RCNN variant to improve performance during extreme events, which may be absent from training data.

The framework efficiently downscales the SST along the west coast of Australia. A case study of the 2011 marine heatwave shows that the RCNN improves ACCESS-S2 SST predictions by increasing the horizontal resolution from 25 km to 2 km, allowing identification of finer-scale anomalies during marine heatwaves that are not resolved in the ACCESS-S2 dataset. By applying the developed framework, this work achieves a balance between computational efficiency and the accuracy required for capturing local oceanic variations, ultimately improving forecasting capabilities for coastal management and marine ecosystem studies.