Modified STARIMA model for space-time data

Abstract

In this paper we propose spatial time series model. ARIMA model class is considered for each location. Model for each location is built by spatial  ``connection'' of identified ARIMA models in observed locations. Spatial  ``connection'' is implemented by spatial averaging of the coefficients of ARIMA models and by ordinary kriging procedure for means. Comparison between proposed model which can be considered as modified STARIMA model and general STARIMA model developed by Pfeifer and Deutch is presented. Mean square prediction error (MSPE) for proposed procedure of prediction is presented. Values of MSPE for real data are calculated.