应用GHSOM网络分析南海北部表层环流模态与黑潮入侵

Variability of surfer circulation and Kuroshio intrusion in northern South China Sea using growing hierarchical self-organizing maps

基于1992年10月至2009年11月卫星观测的海表高度(SSH)时间序列数据,应用增长型分级自组织映射(GHSOM)人工神经网络方法研究南海北部和西太平洋SSH和中尺度涡旋的变化,识别出该海域SSH的季节和年际变化信号。分析表明,流经吕宋海峡的黑潮分支在冷季入侵南海北部,同时在吕宋岛西北海域出现一个强烈的气旋式涡旋,表层黑潮的入侵与跨过吕宋海峡南北的经向压力梯度密切相关。黑潮的非入侵事件主要出现在暖季。春秋季节作为两个事件的过渡期,环流结构复杂,由GHSOM的第2层特征图进一步进行分类识别。黑潮入侵事件和非入侵事件发生的百分比分别为24.57%和27.53%,过渡模态的百分比为47.87%。当入侵南海事件发生时,南海北部表层环流流态相对简单,主要为气旋环流控制南海北部,吕宋海峡表层海流是否入侵南海,与南海北部中尺度涡旋特别是吕宋岛西北的气旋式涡的变化关系密切;反之,在非入侵事件发生时,南海北部出现多涡结构,环流流态复杂,表明吕宋海峡海流入侵南海对南海北部环流也有重要调整作用。除季节尺度变化外,年际时间尺度变化信号也十分显著。在1994—1995、1997—1998和2002—2003年期间,表层黑潮入侵南海北部的事件要显著多于其他年份,然而入侵事件在1998—2001年和2006—2009年时间段明显减少,非入侵事件增加。应用欧氏距离定义的模态2的时间发展序列与Ni o3.4指数序列延迟相关。

Sea surface height （SSH） variations and eddies on both sides of the Luzon Strait are examined from the merged satellite altimeter data from October 1992 to November 2009. The neural network analyses based on the growing hierarchical self-organizing map （GHSOM） are used to extract feature patterns of the circulation variability. The evolution of the characteristic circulation patterns with time is investigated. The seasonal and inter-annual variations of the sea-surface circulation are identified. The analysis indicates a branch of the Kuroshio intrudes into the northern South China Sea （NSCS） during the cold season, accompanied by a stronger cyclonic eddy northwest of Luzon Island. That is closely related to the meridional pressure gradient cross the Luzon Strait. The Kuroshio does not intrude remarkably in the warm season. The spring and autumn, as transitional periods, have relatively complicated circulation structures, which can be explained by more GHSOM arrays in the second layer. The prominent Kuroshio intrusion and non-intrusion patterns denote 24.57%and 27.53%of all SSH variation, respectively. The transitional pattern sums up to 47.87%. The seasonal variation is modulated by the inter-annual variation. The events that the surface Kuroshio intruded into the NSCS in 1994？1995, 1997？1998 and 2002？2003 are much more frequent than the other periods, and the frequency of the events decreased remarkably from 1998 to 2001 and from 2006 to 2009, with no intrusion events were prominent in these two periods. The time series of non-intrusion pattern is best correlated with the Ni？o3.4 index, when it lags the Ni？o3.4 index by one month.