Analysis of Rainfall Prediction Using Parallel Hybrid Algorithm

Abstract

Precisely forecasting rainfall precipitation is an intricate and crucial challenge faced by numerous weather forecasters. In this study, we conducted an examination of different statistical models to assess their efficacy in predicting monthly rainfall precipitation. The objective of this study was to develop a combined model that could enhance the accuracy of such forecasts. To achieve this, we gathered monthly rainfall time series data spanning from January 1901 to December 2017 in Tamil Nadu, India. To enhance the accuracy of rainfall precipitation prediction, we employed a parallel hybrid strategy, combining univariate forecast models. Our proposed forecasting model was compared with other established models, including Seasonal Auto-Regressive Integrated Moving Average (SARIMA), Holt-Winters Additive (HWA) model, Holt model, Exponential Smoothing (ETS) model, and Feed Forward Neural Network (FFNN) model. The results indicate that our proposed model outperformed the other models, demonstrating its superior forecasting capabilities. The proposed model yielded an RMSE value of 0.6403, MSE value of 0.4101, MAE value of 0.3998, NSE value of 0.5924, sMAPE value of 0.7172, and an R-value of 0.7761. A paired t-test was conducted to compare the performance metrics of the proposed model with those of the baseline models. The result shows that this model is statistically significant. Since, It p-value less than 0.05. These findings lead us to the conclusion that the proposed model is well-suited for analyzing and forecasting climatological factors and climatic extremes.