Modelling of non-stationary spatio-temporal data with estimation, testing and forecasting

2016 Fiscal Year Final Research Report

• Project/Area Number: 25280005
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Partial Multi-year Fund
• Section: 一般
• Research Field: Statistical science
• Research Institution: Tohoku University
• Principal Investigator: Matsuda Yasumasa 東北大学, 経済学研究科(研究院), 教授 (10301590)
• Co-Investigator(Kenkyū-buntansha): 陳 春航 琉球大学, 理学部, 准教授 (00264466) ; 栗原 考次 岡山大学, その他の研究科, 教授 (20170087) ; 柿沢 佳秀 北海道大学, 経済学研究科(研究院), 教授 (30281778) ; 西山 慶彦 京都大学, 経済研究所, 教授 (30283378) ; 丸山 祐造 東京大学, 学内共同利用施設等, 准教授 (30304728) ; 生川 雅紀 岡山大学, 社会文化科学研究科, 准教授 (30588489) ; 西井 龍映 九州大学, 学内共同利用施設等, 教授 (40127684) ; 高橋 邦彦 名古屋大学, 医学(系)研究科(研究院), 准教授 (50323259) ; 矢島 美寛 東京大学, 経済学研究科(研究院), 教授 (70134814)
• Project Period (FY): 2013-04-01 – 2017-03-31
• Keywords: spatio-temporal data / Whittle estimate / spectral density / kriging / forecasting / separable correlations / irregularly spaced data

Outline of Final Research Achievements

We have developed a model to analyze huge spatio-temporal data set with non-stationary structures and proposed a method of parameter estimation, testing of correlation structures and forecasting with applications to empirical data mainly from environmental studies. Specifically, We have considered a separable model whose spatio-temporal correlations are given by product of spatial and temporal correlations. In the empirical studies in this research, we employed continuous autoregressive moving average (CARMA) models for spatial behaviors and traditional ARMA models for temporal ones. We applied the separable models to the US rain fall data recorded monthly at 6000 points irregularly spaced inside US continent and found proposed an efficient estimation, testing and forecasting methods that can cope with huge data set composed of several million data points.

Free Research Field

時系列解析、時空間統計学