时空团状系数回归模型及其在海洋时空断面数据中的应用

Spatio-Temporally Clustered Coefficient Model and Its Application in Ocean Spatio-Temporal Section Data

为揭示变量间的回归关系在时空域上的变化结构,本文提出了一种新的变系数回归模型——时空团状系数(Spatio-temporally clustered coefficient, STCC)回归模型。STCC模型通过对时空上相邻点的回归系数之差施加惩罚,从而估计出存在时空变化的回归关系。该模型可在没有先验信息的情况下探索回归关系的时空结构。数值模拟实验表明:STCC模型不仅能有效捕捉回归关系在时空域上的团状结构,对随时空连续变化的回归系数也表现出了较好的估计性能。本文运用STCC模型探索了大西洋25°W断面上的海水温度和盐度之间的回归关系,并据此初步分析了南极中层水的季节性演变特征。

To reveal the spatial and temporal structure of relationship between variables,we propose the spatio-temporally clustered coefficient(STCC)regression model.The STCC model punishes the differences between adjacent regression coefficients to encourage the spatio-temporal homogeneity of coefficients.Numerical simulations show that the STCC not only effectively capture the cluster in regression coefficients,but also exhibits good estimation performance when regression coefficients vary continuously over time and space.We apply the STCC model to exploring the relationship between seawater temperature and salinity on the 25°W longitude section in the Atlantic Ocean,and preliminarily analyze the seasonal evolution characteristics of Antarctic intermediate water.