Sea Surface Temperature Forecasting During Extreme Weather Using a Multi-variable Patch-Based Deep Learning Framework

Dat Nguyen Khanh

doi:10.1016/j.icte.2026.08.007

Sea Surface Temperature Forecasting During Extreme Weather Using a Multi-variable Patch-Based Deep Learning Framework

2026-08-27

0

OA

AI

DOI：10.1016/j.icte.2026.08.007

Original

Original request for help

Save

Abstract

En 中文

Accurate sea surface temperature (SST) forcasting is essential for ocean monitoring, especially during extreme weather events when air-sea interactions intensify. This study introduces TCN-PatchTST, an enhanced variant of PachTST, utilizing observations from four buoys around Taiwan for 72-hour SST forecasting. The results indicate that TCN-PatchTST consistently outperforms physical and data-driven baselines, including HYCOM, XGBoost, LSTM and the standard PatchTST in terms of prediction accuracy and demonstrates greater stability during extreme weather conditions. Additionally, SHAP analysis interprets the main factors influencing SST variability. Findings suggest that this multivariable approach significantly improves the reliability of SST forecasts across varying environmental conditions.

Keywords:

Explainable AI (SHAP)

Forecasting

Patch-based deep learning

Sea surface temperature

Typhoon

Sea Surface Temperature Forecasting During Extreme Weather Using a Multi-variable Patch-Based Deep Learning Framework

Abstract

Journal

Organization