Forecasting seasonal rainfall with time series, machine learning and deep learning

Abstract

 Seasonal rainfall forecasting is crucial for agricultural planning and water resource management in Delta State, Nigeria, as the region's economy is highly dependent on climate. This study investigates the trend and appropriate models for forecasting seasonal rainfall patterns in the region. We employed a range of methods, including traditional time series techniques like Holt's Winter exponential smoothing and the Seasonal Autoregressive Integrated Moving Average (SARIMA), alongside more advanced machine learning and deep learning models. Critical data properties such as stationarity and normality of error terms were first assessed. Model performance was then evaluated using standard metrics, including root mean square error, mean absolute error, mean absolute percentage error, and mean square error. The data was found to have a stationary pattern, and among the models explored, the Holt's Winter exponential smoothing model was identified as the best performing.