Rainfall Prediction Using Gated Recurrent Unit Based on DMI and Nino3.4 Index

Rainfall variability has a severe impact in Sidoarjo, Indonesia. The significant increase in extreme rainfall caused a hydrometeorological disaster and expanded the Sidoarjo mudflow. Rainfall prediction can reduce risks and anticipate hydrometeorological disasters. This study predicts rainfall based on the Dipole Mode Index (DMI) and Niño3.4 Index and several other parameters such as temperature, humidity, duration of sunshine, and wind speed. This study uses monthly time-series data to predict rainfall and compare the results of the 1D-CNN, RNN, LSTM, and GRU methods. The best prediction was made by GRU with a Mean Arctangent Absolute Percentage Error (MAAPE) value of 0.42 and R-square value of 0.79 with 32 hidden neurons, 32 batch sizes, and 0.001 learning rate. Predictions indicate that the rainfall intensity will increase from 50 mm to 200 mm per month from September 2021 to January 2022, or the rainfall intensity will increase by 30 mm per month.