SARIMA MODELLING OF MONTHLY RAINFALL IN RIVERS STATE OF NIGERIA
1Love Cherukei Nnoka , 2Ifeoma Better Lekara-Bayo & 2Ette Harrison Etuk
1Department of Statistics, Elechi Amadi Polytechnic, Rumuola, Port Harcourt, Nigeria
2Department of Mathematics, Rivers State University, Port Harcourt, Nigeria
Email: nnokalove@yahoo.com
ABSTRACT
This work is about seasonal autoregressive integrated moving average (SARIMA) modeling of monthly rainfall of Rivers State of Nigeria from 1981 to 2016. The time plot shows seasonality of period 12 months as typical of rainfall data. Even though the Augmented Dickey Fuller test of unit root ceritifies the series as stationary the correlogram shows an undulating sinusoidal pattern of seasonality of period 12, as expected. The correlogram shows positive spikes at lag 12 and comparable spikes at the autocorrelation function (ACF) at lags 11 and 13 and spikes at lags 12 and 24 on the partial autocorrelation function (PACF). This suggests the involvement of a seasonal autoregressive order of 1 or 2 and a moving average non-seasonal order of 1 and a seasonal order of 1. This means the involvement of a SARIMA(0,0,1)x(1,1,1)12 model. Other models worth testing are SARIMA(0,0,1)x(2,1,1)12 model and a SARIMA(1,0,1)x(0,1,1)12 model. By AIC and R2 the lattest model was chosen. The correlogram of the residuals showed no significant spike, an evidence of model adequacy. The forecasts of 2017 were obtained finally.
Keywords: Rainfall, Rivers State, Nigeria, SARIMA modeling