Space-Time ARIMA Modeling for Regional Precipitation Forecasting

Abstract

An aggregate regional forecasting model class belonging to the general family of space-time auto regressive moving average (STARMA) process is investigated. These models are characterised by autoregressive and moving average terms lagged in both time and space. The paper demonstrates an iterative procedure for buling a starima model of precipitation time series. Eleven raingage sites located in a watershed in southern Ontario, Canada, sampled at 15-day intervals for the period of 1966 to 1980 are used in the numerical analysis. The identified model is STMA($l_2)$. The model parameters are estimated by the polytope technique, a nonlinear optimization algorithm. The developed model performed well in regional forecasting and in describing the spatio-temporal characteristics of the precipitation time series.