Predicting rainfall patterns in Palakkad district of South India based on time series forecasting approaches

Abstract

Rainfall prediction is inevitable for managing agricultural activities, water resources, and mitigating the risks of droughts and floods, particularly in regions like the Palakkad district, Kerala, which is known for its agricultural significance. Time series forecasting plays a vital role in identifying patterns, trends, and seasonal variations in rainfall data. In this study the Holt-Winters exponential smoothing model and the seasonal autoregressive integrated moving average (SARIMA) model are considered for predicting the rainfall in Palakkad district, Kerala. The forecast accuracy metrics root mean squared error (RMSE) and mean absolute percent error (MAPE) are utilised to evaluate the performance of the above models. While the Holt-Winters exponential smoothing model effectively captured simple seasonal patterns, the SARIMA model excelled in handling complex seasonal structures and trends. The SARIMA model was identified as the most accurate, with minimal forecast errors, adherence to assumptions, and low residual correlations.