Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Leveraging an Atmospheric Foundational Model for Subregional Sea Surface Temperature Forecasting

Víctor Medina, Giovanny A. Cuervo-Londoño, Javier Sánchez

TL;DR

The paper tackles the challenge of accurate, scalable SST forecasting in a subregional upwelling system by repurposing the Aurora atmospheric foundation model for ocean data. It fine-tunes a 3D Transformer-based architecture on high-resolution ocean reanalysis (GLORYS12V1) to predict potential temperature in the Canary region, employing a two-phase fine-tuning strategy and latitude-weighted evaluation. The approach achieves RMSE around 0.12–0.14 K with an anomaly correlation close to 0.997, capturing large-scale SST structures while noting coastal-region limitations and the need for higher resolution and additional variables. This cross-domain transfer demonstrates the potential of foundational models to reduce computational costs and enable rapid experimentation in ocean forecasting, with future work centered on physics-informed training, multi-variable integration, and continual learning to enhance interpretability and accuracy.

Abstract

The accurate prediction of oceanographic variables is crucial for understanding climate change, managing marine resources, and optimizing maritime activities. Traditional ocean forecasting relies on numerical models; however, these approaches face limitations in terms of computational cost and scalability. In this study, we adapt Aurora, a foundational deep learning model originally designed for atmospheric forecasting, to predict sea surface temperature (SST) in the Canary Upwelling System. By fine-tuning this model with high-resolution oceanographic reanalysis data, we demonstrate its ability to capture complex spatiotemporal patterns while reducing computational demands. Our methodology involves a staged fine-tuning process, incorporating latitude-weighted error metrics and optimizing hyperparameters for efficient learning. The experimental results show that the model achieves a low RMSE of 0.119K, maintaining high anomaly correlation coefficients (ACC $\approx 0.997$). The model successfully reproduces large-scale SST structures but faces challenges in capturing finer details in coastal regions. This work contributes to the field of data-driven ocean forecasting by demonstrating the feasibility of using deep learning models pre-trained in different domains for oceanic applications. Future improvements include integrating additional oceanographic variables, increasing spatial resolution, and exploring physics-informed neural networks to enhance interpretability and understanding. These advancements can improve climate modeling and ocean prediction accuracy, supporting decision-making in environmental and economic sectors.

Leveraging an Atmospheric Foundational Model for Subregional Sea Surface Temperature Forecasting

TL;DR

The paper tackles the challenge of accurate, scalable SST forecasting in a subregional upwelling system by repurposing the Aurora atmospheric foundation model for ocean data. It fine-tunes a 3D Transformer-based architecture on high-resolution ocean reanalysis (GLORYS12V1) to predict potential temperature in the Canary region, employing a two-phase fine-tuning strategy and latitude-weighted evaluation. The approach achieves RMSE around 0.12–0.14 K with an anomaly correlation close to 0.997, capturing large-scale SST structures while noting coastal-region limitations and the need for higher resolution and additional variables. This cross-domain transfer demonstrates the potential of foundational models to reduce computational costs and enable rapid experimentation in ocean forecasting, with future work centered on physics-informed training, multi-variable integration, and continual learning to enhance interpretability and accuracy.

Abstract

The accurate prediction of oceanographic variables is crucial for understanding climate change, managing marine resources, and optimizing maritime activities. Traditional ocean forecasting relies on numerical models; however, these approaches face limitations in terms of computational cost and scalability. In this study, we adapt Aurora, a foundational deep learning model originally designed for atmospheric forecasting, to predict sea surface temperature (SST) in the Canary Upwelling System. By fine-tuning this model with high-resolution oceanographic reanalysis data, we demonstrate its ability to capture complex spatiotemporal patterns while reducing computational demands. Our methodology involves a staged fine-tuning process, incorporating latitude-weighted error metrics and optimizing hyperparameters for efficient learning. The experimental results show that the model achieves a low RMSE of 0.119K, maintaining high anomaly correlation coefficients (ACC ). The model successfully reproduces large-scale SST structures but faces challenges in capturing finer details in coastal regions. This work contributes to the field of data-driven ocean forecasting by demonstrating the feasibility of using deep learning models pre-trained in different domains for oceanic applications. Future improvements include integrating additional oceanographic variables, increasing spatial resolution, and exploring physics-informed neural networks to enhance interpretability and understanding. These advancements can improve climate modeling and ocean prediction accuracy, supporting decision-making in environmental and economic sectors.

Paper Structure

This paper contains 18 sections, 11 equations, 9 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Dataset and Study Area
  4. Global Ocean Physics Reanalysis Data
  5. Area and Period of Study
  6. The Potential Temperature
  7. A Foundational Model for Forecasting the SST
  8. The Encoder
  9. The Processor
  10. The Decoder
  11. Subregional Oceanographic Prediction
  12. Experimental Configuration
  13. Dataset Split
  14. Loss Function
  15. Evaluation Metrics
  16. ...and 3 more sections

Figures (9)

  • Figure 1: Study area showing the African coast, the Canary Islands, and relevant capes.
  • Figure 2: Spatial representations of the dataset variables from the GLORYS12V1 product.
  • Figure 3: Temporal series of the study period for the $\theta_0$ variable.
  • Figure 4: Prediction of the pre-trained Aurora model with meteorological data after reducing data resolution to $0.5^\circ$.
  • Figure 5: Dataset split into the training, validation, and test sets.
  • ...and 4 more figures