徐为帅, 张磊, 王华, xxxx. 基于锋线提取和VMD-LSTM的东海黑潮温度锋强度预测[J]. 海洋科学进展, x(x): xx-xx. doi: 10.12362/j.issn.1671-6647.20230606001.
引用本文: 徐为帅, 张磊, 王华, xxxx. 基于锋线提取和VMD-LSTM的东海黑潮温度锋强度预测[J]. 海洋科学进展, x(x): xx-xx. doi: 10.12362/j.issn.1671-6647.20230606001.
XU W S, ZHANG L, WANG H, xxxx. Prediction of temperature front strength of Kuroshio in the East China Sea based on front extraction and VMD-LSTM[J]. Advances in Marine Science, x(x): xx-xx. doi: 10.12362/j.issn.1671-6647.20230606001
Citation: XU W S, ZHANG L, WANG H, xxxx. Prediction of temperature front strength of Kuroshio in the East China Sea based on front extraction and VMD-LSTM[J]. Advances in Marine Science, x(x): xx-xx. doi: 10.12362/j.issn.1671-6647.20230606001

基于锋线提取和VMD-LSTM的东海黑潮温度锋强度预测

Prediction of Temperature Front Strength of Kuroshio in the East China Sea Based on Front Extraction and VMD-LSTM

  • 摘要: 东海黑潮锋是黑潮与东海陆架水的交界,对东海海洋环境和气候变化产生重要影响。本文基于JCOPE2M(Japan Coastal Ocean Predictability Experiment 2 Modified)再分析数据和WOA23(World Ocean Atlas 2023)气候态平均数据,结合等温线和纬向最高温度梯度提取东海黑潮锋的锋线,针对东海黑潮锋强度存在不同时间尺度变化的特性,构建了基于变分模态分解(Variational Mode Decomposition, VMD)和长短期记忆网络(Long Short-Term Memory, LSTM)的预测模型,研究结果表明:本研究提出的锋线识别方法适用于全水深和弱锋区域,在东海黑潮锋识别中取得了较好的效果;通过锋线提取得到的东海黑潮锋强度序列可分解为趋势变化项、周期变化项及不规则波动变化项,次表层水锋强显著高于表层和中层水,并呈逐年递增的趋势,可达0.033 4 (℃/km)/a;VMD-LSTM组合模型预测的东海黑潮锋强度与原始值的接近程度高,预测效果优秀,精度较季节性差分自回归滑动平均模型和LSTM模型提升显著,具有一定的推广价值。

     

    Abstract: The Kuroshio front in the East China Sea is the boundary between the Kuroshio Current and the East China Sea shelf water, exerting significant influence on the marine environment and climate variability in the region. In this study, based on JCOPE2M reanalysis data and WOA23 climatological average data, the Kuroshio front is identified using isotherms and the meridional maximum temperature gradient. The seasonal and interannual variations of the Kuroshio front strength are analyzed statistically. Considering the multi-scale characteristics of the Kuroshio front, a prediction model for the Kuroshio front in the East China Sea is constructed with the Variational Mode Decomposition (VMD) and the Long Short-Term Memory (LSTM) networks. The results indicate that the proposed method for front detection and optimization is suitable for both the entire water column and weak front regions, achieving satisfactory results in identifying the Kuroshio front in the East China Sea. The strength sequence of the Kuroshio front extracted through frontal analysis in the East China Sea can be decomposed into trend, periodic, and irregular fluctuation components. The sub-surface front exhibits significantly higher strength compared to the surface and mid-layer fronts, with an increasing trend of 0.033 4 (℃/km)/a. The VMD-LSTM combined model shows excellent predicting performance, with higher accuracy and better agreement with the original values of the Kuroshio front strength. The prediction accuracy is significantly improved compared to the seasonal autoregressive integrated moving average model and the standalone LSTM model, indicating its potential for broader applications.

     

/

返回文章
返回