Scholarlink

<p>Accurate large-scale ocean environment modeling is challenged by multi-scale dynamics, sparse sampling, and measurement noise. This paper presents a fast spatiotemporal (ST) modeling and robust target diagnosis for large-scale ocean environments. First, the four-dimensional field is factorized via a Karhunen-Loève (KL) expansion into orthonormal spatial modes and their temporal features. The spatial modes are further regularized by smooth parametric functions, while the temporal features follow a compact nonlinear evolution, enabling efficient ST fusion for reconstruction. Building on the denoised field, the diagnosis module applies depth-wise Savitzky-Golay smoothing, prominence-based peak search on the vertical temperature gradient, and a half-maximum rule to estimate thermocline depth and thickness. Experiments on a Pacific Ocean dataset demonstrate favorable efficiency, stability, and interpretability. The proposed method achieves a root mean square error 0.1945° C in temperature reconstruction tests, while delivering reliable thermocline localization and thickness estimation suitable for online deployment.</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

Fast Spatiotemporal Modeling and Robust Target Diagnosis for Large-Scale Ocean Environments

  • 期刊论文

  • 发表日期：

    2026年4月28日

查看全文
  • Lei Lei,
  • Xi Chen,
  • Ben M. Chen

摘要

Accurate large-scale ocean environment modeling is challenged by multi-scale dynamics, sparse sampling, and measurement noise. This paper presents a fast spatiotemporal (ST) modeling and robust target diagnosis for large-scale ocean environments. First, the four-dimensional field is factorized via a Karhunen-Loève (KL) expansion into orthonormal spatial modes and their temporal features. The spatial modes are further regularized by smooth parametric functions, while the temporal features follow a compact nonlinear evolution, enabling efficient ST fusion for reconstruction. Building on the denoised field, the diagnosis module applies depth-wise Savitzky-Golay smoothing, prominence-based peak search on the vertical temperature gradient, and a half-maximum rule to estimate thermocline depth and thickness. Experiments on a Pacific Ocean dataset demonstrate favorable efficiency, stability, and interpretability. The proposed method achieves a root mean square error 0.1945° C in temperature reconstruction tests, while delivering reliable thermocline localization and thickness estimation suitable for online deployment.