Rainfall Forecasting for Odeda, South-Western Nigeria using a Seasonal ARIMA Time-Series Model

Rainfall occurrence within Odeda in Southwestern Nigeria. For this purpose, rainfall records spanning from January 1982 to December 2016 of the Ogun-Oshun River Basin Development Authority weather station were used to develop the model. The box-Jenkins approach was adopted in which tentative models were identified by the visual inspection of the charts of the autocorrelation function (ACF) and partial autocorrelation (PACF) of each model after subjecting the raw data to seasonal differencing of order 1 and ordinary differencing of order 1 to eliminate seasonality. The coefficient of determination (R2), Bayesian information criterion (BIC), and Root Mean Square Error (RMSE) were used also used to select preliminary models as criteria before each model selected was subjected to diagnostic checking by inspection of the ACF and PACF charts for noise residuals, histogram of noise residuals for normality, and normal Q-Q plot of noise residuals. The best model was observed to be a Seasonal ARIMA (3,1,3) (2,1,1)12 model having satisfied all the selection criteria. This model was used to validate a forecasted rainfall event between January 2017 and December 2018 and was found to be adequate with a coefficient of determination of 0.68 when compared with the actual rainfall events for that period. The model was used further to forecast rainfall events up to 2022.