冬春季黄海温度锋面的多时间尺度变化及主控因素分析

海洋锋面强度变化对陆源物质输运和全球物质循环有重要作用。冬春季节,中国东部陆架区西太平洋边界流分支与沿岸流之间形成了海洋温度锋。为探究冬季风暴和陆架环流双重影响下温度锋面的多时间尺度变化及主控因素,本文以黄海为研究区,分别在年代际尺度和天气尺度,利用信号分解和可解释深度学习方法,研究了低纬度驱动的环流系统和高纬度驱动的冬季风暴对锋面变化的耦合作用。在年代际尺度,通过使用经验正交函数分解和集合经验模态分解的方法,将北黄海的温度变化与黄海暖流强度相联系。研究结果表明,黄海的海表温度经验正交函数（EOF）分解第一模态的空间分布有明显的黄海暖流—沿岸流体系特征;海表温度EOF第一模态时间序列与黄海暖流强度指标的相关性良好,且受低频率厄尔尼诺‒南方涛动信号调控。在天气尺度,对卷积神经网络−长短时记忆网络（CNN-LSTM）模型进行训练并使用可解释性指标进行分析,结果发现无风或弱风条件下,海洋锋面主要由压力梯度力和科里奥利力的地转平衡维持;但在冬季风暴条件下,受开尔文波传播和切变锋破碎的影响,流场的低频波动成为导致锋面强度变化的主因。本文研究结果表明,大数据及机器学习方法是在众多海洋参数间建立联系,并发现一些独特物理海洋过程的重要手段,具有广阔的应用前景。

Multi-timescale variation of temperature fronts in the Yellow Sea during winter and spring and its main controlling factors analysis

Ocean fronts variations in strength are key to the terrestrial material transport and global material cycle. Ocean temperature fronts are formed between the branches of West Pacific Boundary Current and the coastal current during the winter and spring seasons in the eastern shelf of China. In order to investigate the multi-time scale variation and main controlling factors of temperature front over the Yellow Sea under the dual influence of winter storms and shelf circulation, we investigate the coupling of low-latitude driven circulation systems and high-latitude driven winter storms on frontal variability with the methods of signal decomposition and explainable deep learning on the decadal and weather scale. On the decadal scale, empirical orthogonal decomposition and ensemble empirical modal decomposition are used to relate temperature changes in the Yellow Sea to the strength of the Yellow Sea Warm Current. The results indicate that the spatial distribution of first sea surface temperature (SST) EOF mode has obvious characteristics of the Yellow Sea Warm Current-coastal current system; the time series of the first SST EOF mode correlates well with the Yellow Sea Warm Current intensity index and is modulated by the low frequency ENSO signal. On the weather scale, this paper trains CNN-LSTM models and uses interpretability metrics to conduct the research. The results show that, in windless or weak wind conditions, the strength of ocean front is maintained by the combination of pressure gradient forces resulted from sea surface height and Coriolis forces caused by flow field. However, in the storm conditions, influenced by Kelvin Wave propagation and shear front fragmentation, the flow field is responsible for the ocean front variation. The results of this study show that big data and machine learning methods are important means to establish connections between many ocean parameters and discover some unique physical ocean processes, which have broad application prospects.