Advances in Biology & Earth Sciences

This study develops an extreme weather prediction model in the Maritime Continent using a hybrid approach combining Artificial Neural Networks (ANN) and Reservoir Computing (RC). The datasets used include Madden-Julian Oscillation (MJO), weather and rainfall data, which are processed by calculating the coordinate centroid for each location. Two ANN models are developed for prediction: temperature prediction for the weather dataset and rainfall prediction for the rainfall dataset. The methodology of this study involves several stages; first, a comprehensive dataset is integrated into MJO, weather and rainfall data. Second stages is the weather and rainfall datasets that are processed by calculating the centroid location and removing missing data. The ANN model is trained using the processed dataset and integrated with MJO data. The training results show the best validation performance for Weather-MJO at epoch 797, with a value of 4.845e-08 and for Rain-MJO at epoch 83, with a value of 0.00035924. In addition, other performance values ​​show promising results: for Weather-MJO, the gradient reaches 0.0004948 at epoch 803, M_u is 1e-09 at epoch 803, training R is 0.9999, validation R is 0.99998, test R is 0.99381 and all R is 0.99908. Meanwhile, for Rain-MJO, the gradient is 1.1028e-05 at epoch 89, mu is 1e-09 at epoch 89, training R is 0.065633, validation R is 0.058097, test R is 0.061123 and all R is 0.063837. This model can improve extreme weather prediction in the Maritime Continent and support disaster risk management.