AI Boosts Sea Temperature Data for Improved Forecasts

Abstract

Diurnal variability in sea surface temperature (SST) significantly influences ocean-atmosphere thermal interactions. Conventional numerical methods for reconstructing hourly SST are limited by high computational demands and difficulties in accommodating real-time atmospheric and oceanic conditions. Therefore, this study introduces the Physics-Assisted Reconstruction Adversarial Network (PARAN), a novel deep-learning framework that reconstructs high-resolution (2 km), hourly SST using geostationary satellite data. By integrating numerical model knowledge and leveraging generative adversarial network architectures, PARAN achieves enhanced reconstruction performance. We applied the PARAN framework to the Northwest Pacific region, generating a continuous, hourly subskin SST layer from January 2019 to December 2021. The network was validated against high-resolution satellite data and in situ buoy observations, demonstrating substantial accuracy improvements over existing high-resolution numerical models (GLO12v4 and HYCOM). We obtained higher correlation coefficients (0.994 and 0.982), negligible biases (0.007 and −0.165), and lower root mean square errors (0.435 and 0.766) than those of existing models when comparing PARAN with drifting and mooring buoys, respectively. In particular, PARAN effectively captured detailed spatiotemporal SST variations, including diurnal warming, and showed robustness under both clear and cloudy conditions.

Every summer, typhoons threatening the Korean Peninsula draw their energy from the warm waters of the Northwest Pacific Ocean. In recent years, the frequency and intensity of extreme weather events-such as heatwaves, droughts, and heavy rains-have been increasingly linked to rising sea surface temperatures (SST). Accurate prediction of SSTs has thus become a vital component of climate and weather forecasting. However, satellite observations, which provide broad and continuous monitoring, often suffer from data gaps caused by clouds, precipitation, and other observational limitations, hampering long-term, high-resolution climate analysis. Responding to this challenge, a team of researchers at UNIST has developed a pioneering artificial intelligence (AI) model capable of restoring missing satellite data and generating continuous, high-resolution SST datasets with unprecedented accuracy.

Led by Professor Jungho Im from the Department of Civil, Urban, Earth, and Environmental Engineering, the team announced that they have created an innovative AI-based reconstruction system that fills in observational gaps, producing SST data at a 2-kilometer spatial resolution and on an hourly basis. This development promises to significantly enhance our understanding of oceanic conditions that directly influence regional weather and climate patterns.

The ocean retains approximately 90% of Earth's surface energy, with SST serving as a critical boundary where heat exchange between the ocean and atmosphere occurs. Elevated SSTs can transfer heat upward, fueling powerful typhoons, intensifying heatwaves, and increasing the risk of heavy rainfall events. Yet, despite its importance, continuous, high-resolution monitoring of SST remains challenging due to data gaps in satellite imagery.

To overcome this obstacle, the researchers employed a Generative Adversarial Network (GAN)-a sophisticated AI architecture originally designed for image synthesis-and trained it using high-frequency satellite data combined with thermodynamic insights from numerical weather prediction (NWP) models. Unlike conventional models, this approach integrates physical oceanic principles, enabling the AI to produce SST data that aligns closely with real-world physical conditions, even in the presence of missing observations.

"Traditional methods like linear interpolation or statistical models often struggle to preserve the fine details of SST, especially during rapid temperature changes," explained Sihun Jung, the study's first author. "Our AI model not only surpasses these methods in accuracy but also maintains high fidelity even in challenging conditions, making it a powerful tool for climate monitoring."

Professor Im emphasized the broader impact, saying "This advanced reconstruction technology is particularly crucial for the Northwest Pacific, a region prone to frequent typhoons and climate variability." He further noted, "By providing high-resolution SST data, we can significantly improve weather forecasts and climate models. In the long run, this technology could also be instrumental in early warning systems for marine disasters, such as marine heatwaves, helping to safeguard communities and ecosystems."

The findings of this research have been published in Remote Sensing of Environment (Impact Factor: 11.1) on June 1, 2025. Recognized globally as a leading publication in remote sensing and environmental sciences, this journal underscores the significance of the study's contribution to advancing satellite-based environmental monitoring and climate prediction technologies.

Journal Reference

Sihun Jung, Jungho Im, Daehyeon Han, "PARAN: A novel physics-assisted reconstruction adversarial network using geostationary satellite data to reconstruct hourly sea surface temperatures," Remote Sens. Environ., (2025).