Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SEA-ViT: Sea Surface Currents Forecasting Using Vision Transformer and GRU-Based Spatio-Temporal Covariance Modeling

Teerapong Panboonyuen

TL;DR

SEA-ViT tackles forecasting of sea surface currents using high-frequency radar by marrying Vision Transformer (ViT) with bidirectional GRUs to model spatio-temporal covariance in $U$ and $V$. The approach processes a rich 30-year HF radar dataset and incorporates the ENSO index to capture climate-driven variability, while enforcing physical constraints from fluid dynamics through a physics-informed loss. The architecture blends temporal memory via BiGRU with spatial self-attention, augmented by data normalization and augmentation, and deployed through an MLOps pipeline with a Swagger UI for real-time predictions. The work demonstrates a practical pathway for accurate, climate-aware current forecasting in Thai waters (Gulf of Thailand and Andaman Sea), with direct applicability to maritime operations and environmental monitoring within GISTDA’s framework.

Abstract

Forecasting sea surface currents is essential for applications such as maritime navigation, environmental monitoring, and climate analysis, particularly in regions like the Gulf of Thailand and the Andaman Sea. This paper introduces SEA-ViT, an advanced deep learning model that integrates Vision Transformer (ViT) with bidirectional Gated Recurrent Units (GRUs) to capture spatio-temporal covariance for predicting sea surface currents (U, V) using high-frequency radar (HF) data. The name SEA-ViT is derived from ``Sea Surface Currents Forecasting using Vision Transformer,'' highlighting the model's emphasis on ocean dynamics and its use of the ViT architecture to enhance forecasting capabilities. SEA-ViT is designed to unravel complex dependencies by leveraging a rich dataset spanning over 30 years and incorporating ENSO indices (El Niño, La Niña, and neutral phases) to address the intricate relationship between geographic coordinates and climatic variations. This development enhances the predictive capabilities for sea surface currents, supporting the efforts of the Geo-Informatics and Space Technology Development Agency (GISTDA) in Thailand's maritime regions. The code and pretrained models are available at \url{https://github.com/kaopanboonyuen/gistda-ai-sea-surface-currents}.

SEA-ViT: Sea Surface Currents Forecasting Using Vision Transformer and GRU-Based Spatio-Temporal Covariance Modeling

TL;DR

SEA-ViT tackles forecasting of sea surface currents using high-frequency radar by marrying Vision Transformer (ViT) with bidirectional GRUs to model spatio-temporal covariance in and . The approach processes a rich 30-year HF radar dataset and incorporates the ENSO index to capture climate-driven variability, while enforcing physical constraints from fluid dynamics through a physics-informed loss. The architecture blends temporal memory via BiGRU with spatial self-attention, augmented by data normalization and augmentation, and deployed through an MLOps pipeline with a Swagger UI for real-time predictions. The work demonstrates a practical pathway for accurate, climate-aware current forecasting in Thai waters (Gulf of Thailand and Andaman Sea), with direct applicability to maritime operations and environmental monitoring within GISTDA’s framework.

Abstract

Forecasting sea surface currents is essential for applications such as maritime navigation, environmental monitoring, and climate analysis, particularly in regions like the Gulf of Thailand and the Andaman Sea. This paper introduces SEA-ViT, an advanced deep learning model that integrates Vision Transformer (ViT) with bidirectional Gated Recurrent Units (GRUs) to capture spatio-temporal covariance for predicting sea surface currents (U, V) using high-frequency radar (HF) data. The name SEA-ViT is derived from ``Sea Surface Currents Forecasting using Vision Transformer,'' highlighting the model's emphasis on ocean dynamics and its use of the ViT architecture to enhance forecasting capabilities. SEA-ViT is designed to unravel complex dependencies by leveraging a rich dataset spanning over 30 years and incorporating ENSO indices (El Niño, La Niña, and neutral phases) to address the intricate relationship between geographic coordinates and climatic variations. This development enhances the predictive capabilities for sea surface currents, supporting the efforts of the Geo-Informatics and Space Technology Development Agency (GISTDA) in Thailand's maritime regions. The code and pretrained models are available at \url{https://github.com/kaopanboonyuen/gistda-ai-sea-surface-currents}.
Paper Structure (47 sections, 42 equations, 1 figure, 1 table)

This paper contains 47 sections, 42 equations, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Data Handling and Preprocessing
  3. Data Splitting
  4. Normalization of Sea Surface Current Vectors
  5. Mathematical Rationale
  6. Integration with Temporal and Spatial Features
  7. Model Architecture
  8. Bidirectional GRU Layer
  9. Self-Attention Layer
  10. Training Procedure
  11. Data Overview and Input Format
  12. Detailed Analysis of Input Data
  13. Datetime
  14. Latitude and Longitude
  15. U and V (Sea Surface Current Vectors)
  16. ...and 32 more sections

Figures (1)

  • Figure 1: Proposed GRU-Transformer architecture for predicting sea surface current vectors. This framework is inspired by the transformer-based planning model for symbolic regression presented by Shojaee et al. (2024) shojaee2024transformer.