A transformer-based method for correcting daily SST numerical forecasting products

Frontiers in Earth Science, Journal Year: 2025, Volume and Issue: 13

Published: March 28, 2025

This study introduces applies a Transformer-based method to correct daily Sea Surface Temperature (SST) numerical forecasting products, addressing persistent challenges in short-term SST prediction. The proposed approach utilizes Transformer model architecture capture complex spatiotemporal dependencies error fields, enabling efficient prediction of forecast errors across multiple time scales. was applied hindcast data from the First Institute Oceanography (FIO-COM) ocean system, focusing on northwestern Pacific region. Results demonstrate significant improvements accuracy, with Root Mean Square Error (RMSE) reductions ranging 38.8% for day 2 forecasts 17.6% 5 forecasts. Spatial analysis reveals method’s robust performance diverse oceanographic regimes, including coastal and shelf regions where traditional models often struggle. showed ability reproduce patterns, effectively both large-scale systematic biases smaller-scale regional variations. consistent different horizons suggests potential extending reliable range predictions. findings have important implications applications requiring precise forecasts, operational oceanography, marine weather forecasting, coupled ocean-atmosphere modeling.

Language: Английский

---

A deep learning-based hybrid model for improved SST prediction in the tropical Pacific Ocean

Journal of Oceanology and Limnology, Journal Year: 2025, Volume and Issue: unknown

Published: May 23, 2025

Language: Английский

---

OTCFM: A Sea Surface Temperature Prediction Method Integrating Multi-Scale Periodic Features

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 108291 - 108302

Published: Jan. 1, 2024

Sea surface temperature (SST) is a critical factor in the interaction between ocean and atmosphere, directly influencing global climate patterns dynamic changes marine ecosystems. Accurate prediction of SST great significance for assessing managing change maintaining ecological balance. However, existing methods face challenges such as low accuracy, short periods, significant errors. This paper proposes an innovative deep learning method, Ocean Temperature Cycle Fusion Analysis Model (OTCFM), constructed based on datasets from South China East Sea. approach aims to accurately capture predict cyclical variations variability data provide more precise forecasts temperatures. Firstly, observations SST's seasonal periodic variations, we present partitioning strategy decompose complex into intra-period inter-period variations. Secondly, propose Unit both long-term short-term small-scale changes, moving beyond inherent attributes dataset's frequency time domain characteristics extract feature simultaneously. Finally, by stacking Units using residual connections, alleviate gradient vanishing problem achieve accurate predictions. In this study, with different spatial distribution are selected predictive analysis National Oceanic Atmospheric Administration (NOAA) September 1, 1981, June 7, 2023, total 15,408 data. The experimental results show that OTCFM can evolution temporal processes under conditions. MAE values improved 19.08% 19.52%, respectively, compared convolutional long memory neural network (ConvLSTM), which improves accuracy series has far-reaching impact subsequent promotion sustainable resource management environmental protection.

Language: Английский

---

Deep learning for ocean temperature forecasting: a survey

Intelligent Marine Technology and Systems, Journal Year: 2024, Volume and Issue: 2(1)

Published: Oct. 8, 2024

Abstract Ocean temperature prediction is significant in climate change research and marine ecosystem management. However, relevant statistical physical methods focus on assuming relationships between variables simulating complex processes of ocean changes, facing challenges such as high data dependence insufficient processing long-term dependencies. This paper comprehensively reviews the development latest progress models based deep learning. We first provide a formulaic definition for brief overview learning widely used this field. Using sources model structures, we systematically divide into data-driven physically guided models; explore literature involved each method. In addition, summarize an dataset sea areas, laying solid foundation prediction. Finally, propose current future directions article aims to analyze existing research, identify gaps challenges, complete reliable technical support forecasting, disaster prevention, fishery resource management, promote further research.

Language: Английский

---

Numerical Weather Prediction of Sea Surface Temperature in South China Sea Using Attention-Based Context Fusion Network

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(20), P. 3793 - 3793

Published: Oct. 12, 2024

Numerical weather prediction of sea surface temperature (SST) is crucial for regional operational forecasts. Deep learning offers an alternative approach to traditional numerical general circulation models prediction. In our previous work, we developed a sophisticated deep model known as the Attention-based Context Fusion Network (ACFN). This integrates attention mechanism with convolutional neural network framework. this study, applied ACFN South China Sea evaluate its performance in predicting SST. The results indicate that 1-day lead time, achieves Mean Absolute Error 0.215 °C and coefficient determination (R2) 0.972. addition, situ buoy data were utilized validate forecast results. forecasts using these increased 0.500 corresponding R2 0.590. Comparative analyses show surpasses such ConvLSTM PredRNN terms accuracy reliability.

Language: Английский

---