A nonlinear autoregressive recurrent forward harmonic network for rainfall forecasting

Rainfall plays a crucial role in the Earth’s hydrological cycle. Changes in rainfall patterns adversely affect the climatic conditions and crop productivity. Rain prediction finds application in various domains, such as flood forecasting, agriculture, water management, and so on. Nevertheless, accurate rainfall prediction is regarded as a challenging mission due to the significant variability and intricate characteristics of the atmospheric data. Therefore, the novel technique, namely Nonlinear Autoregressive Recurrent Forward Harmonic Network (NARFH-Net), is developed for forecasting rainfall. Initially, the time series data is gathered from the dataset. After that, the extraction of rainfall indicators is accomplished. Then, feature fusion is performed by the Deep Neural Network (DNN) with Hamming distance. Finally, the developed NARFH-Net is used for the prediction of rainfall. The NARFH-Net is introduced by the integration of Gated Recurrent Unit (GRU), Harmonic Analysis, and Nonlinear Auto-Regressive models with the eXogenous (NARX) model. Here, the GRU can effectively handle sequential data, which captures long-term dependencies and helps the model to maintain complex temporal patterns in time series data. The Harmonic Analysis helps to handle the periodic patterns in rainfall. The NARX can capture the nonlinear relationship between past observations and external influencing variables. This hybrid structure enables the model to effectively learn complex, nonlinear and seasonal characteristics of rainfall data. The In addition, the effectiveness of the presented NARFH-Net is analyzed by utilizing various metrics and the value attained for Mean Absolute Squared Error (MASE) is 0.093, Weighted Absolute Percentage Error (WAPE) is 0.197, Mean Square Error (MSE) is 0.154, Root Mean Square Error (RMSE) is 0.392, Mean Absolute Error (MAE) is 0.253, Mean Absolute Percentage Error (MAPE) is 0.160, Relative Absolute Error (RAE) is 0.250, and R-squared is 0.928.