Bayesian Structural Time Series Model and SARIMA Model for Rainfall Forecasting in Nigeria

Abstract

Nigeria is recognized as being susceptible to climate change, and global warming if not taken care of, will lead to serious problems on livelihoods in Nigeria, especially in the area of agricultural activities. Rainfall is a major determinant of climate change the world over and climate change is one of the foremost global challenge facing humans at the moment. Using monthly time series rainfall data, Bayesian structural time series (BSTS) methodology was applied to fit models through MCMC algorithm. Also, Seasonal Autoregressive Moving Average (SARIMA) models were fitted to the same dataset using Box-Jenkins approach. The two models are considered based on their respective capacities to capture trend, seasonal and structural components of rainfall data. On the basis of model evaluation criteria (RMSE, MAE, MAPE and MASE), the SARIMA model had values that were clearly significantly smaller than that of the BSTS time series model. This implies that the SARIMA model is more robust in its estimations and forecasting abilities. Similarly, the R squared was larger for the SARIMA model than the BSTS (MCMC) model indicating that the SARIMA model was a better fit for the rainfall data. This study shows that SARIMA model is a more precise and robust in dealing with this type of dataset than BSTS (MCMC) model. It is better because its computational process using differencing, lags and moving averages ensure that the underlying components of the model are properly identified and estimated.