Scholarlink

<p>The ocean plays a critical role in regulating global climate and ecosystems, making the accurate prediction of ocean environmental conditions essential for disaster prevention, sustainable resource management, and ecological protection. A novel method for ocean environment prediction has been developed using a Multiscale Spatial-temporal Network (MSSTN), designed to enhance the accuracy of ocean quality forecasting. Spatial and temporal data are leveraged through the integration of deep learning techniques, including Graph Convolutional Networks (GCNs) and attention mechanisms, to capture the complex spatial similarity and temporal dependencies of oceanic events. Data from buoy observations and remote sensing are utilized, and multivariate time series analysis is performed to predict ocean water quality metrics such as chlorophyll a concentration. Quantitative evaluations using Fujian coastal data show that MSSTN reduced the Mean Absolute Percentage Error (MAPE) by 1.035&#xa0;mg/L in 1-day predictions, corresponding to a 12.4% improvement relative to the best existing method, and 1.226&#xa0;mg/L lower MAPE in 1-week forecasts (19.8% improvement over the best existing method). The model sustains an MAPE of &lt; 2.5% in 1-month projections, outperforming conventional methods in both short-term and long-term prediction accuracy and stability.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Ocean environment prediction methods based on deep learning and spatiotemporal feature fusion

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月13日

查看全文
  • Bin Zeng,
  • Rui Wang,
  • Houpu Li

摘要

The ocean plays a critical role in regulating global climate and ecosystems, making the accurate prediction of ocean environmental conditions essential for disaster prevention, sustainable resource management, and ecological protection. A novel method for ocean environment prediction has been developed using a Multiscale Spatial-temporal Network (MSSTN), designed to enhance the accuracy of ocean quality forecasting. Spatial and temporal data are leveraged through the integration of deep learning techniques, including Graph Convolutional Networks (GCNs) and attention mechanisms, to capture the complex spatial similarity and temporal dependencies of oceanic events. Data from buoy observations and remote sensing are utilized, and multivariate time series analysis is performed to predict ocean water quality metrics such as chlorophyll a concentration. Quantitative evaluations using Fujian coastal data show that MSSTN reduced the Mean Absolute Percentage Error (MAPE) by 1.035 mg/L in 1-day predictions, corresponding to a 12.4% improvement relative to the best existing method, and 1.226 mg/L lower MAPE in 1-week forecasts (19.8% improvement over the best existing method). The model sustains an MAPE of < 2.5% in 1-month projections, outperforming conventional methods in both short-term and long-term prediction accuracy and stability.