A novel spatio-temporal outlier detection method within the space-time framework is proposed in this paper.Firstly

a unified framework is developed for constructing spatio-temporal neighborhood

which is based on the space-time statistics and clustering analysis.Then

a spatio-temporal outlier measure involving space-time autocorrelation and heterogeneity is presented.Finally

a tree-step strategy is utilized to detect spatio-temporal outliers.Our method is employed to detect spatio-temporal outliers in Chinese annual temperature database（1970—2002）.A meaningful analysis of the spatio-temporal outliers is also provided.