Exploratory Method for Spatio-Temporal Feature Extraction and Clustering: An Integrated Multi-Scale Framework

ISPRS International Journal of Geo-Information

Wen Luo¹  Zhao-Yuan Yu¹  Sheng-Jun Xiao¹  A-Xing Zhu¹  Lin-Wang Yuan¹ 

[1] Key Lab of Virtual Geographic Environment, Ministry of Education, Nanjing Normal University, Nanjing 210023, China; E-Mails:

关键词: clustering;    scale;    tracking;    oceanography;    classification;    data mining;   

DOI  :  10.3390/ijgi4041870

来源: mdpi

【 摘 要 】

This paper presents an integrated framework for exploratory multi-scale spatio-temporal feature extraction and clustering of spatio-temporal data. The framework combines the multi-scale spatio-temporal decomposition, feature identification, feature enhancing and clustering in a unified process. The original data are firstly reorganized as multi-signal time series, and then decomposed by the multi-signal wavelet. Exploratory data analysis methods, such as histograms, are used for feature identification and enhancing. The spatio-temporal evolution process of the multi-scale features can then be tracked by the feature clusters based on the data adaptive Fuzzy C-Means Cluster. The approach was tested with the global 0.25° satellite altimeter data over a period of 21 years from 1993 to 2013. The tracking of the multi-scale spatio-temporal evolution characteristics of the 1997–98 strong El Niño were used as validation. The results show that our method can clearly reveal and track the spatio-temporal distribution and evolution of complex geographical phenomena. Our approach is efficient for global scale data analysis, and can be used to explore the multi-scale pattern of spatio-temporal processes.

【 授权许可】

CC BY   
© 2015 by the authors; licensee MDPI, Basel, Switzerland.