Artificial Neural Network based Rainfall Prediction using Back Propagation Technique in Pekanbaru city

Abstract

The intensity of rainfall is a significant weather component that has a profound effect on natural disasters, particularly floods and landslides, particularly in Indonesia. Precise meteorological forecasts and comprehensive climatic data, including accurate rainfall projections, will effectively reduce the hazards associated with severe weather events. Prior studies have demonstrated the efficacy of the Backpropagation Neural Network (NN) approach in accurately forecasting rainfall. The objective of this work is to forecast the daily precipitation in Pekanbaru City using Neural Networks with the Backpropagation technique. The neural network model was constructed using supervised multilayer learning, initially with one hidden layer and subsequently expanded to two hidden layers, utilizing daily data spanning three years (2017-2019). The rainfall forecasting model was constructed by many iterative training and testing procedures. Forecasts of rainfall were categorized into six groups: cloudy, light rain, moderate rain, heavy rain, very heavy rain, and extreme rain. The forecast outcomes were shown using a MATLAB graphical user interface (GUI). While the prediction accuracy of 61% falls short of the national verification threshold of 75%, this work establishes a fundamental framework for the application of neural networks in weather forecasting. The outcomes can be enhanced by using more relevant data and using more precise training procedures to achieve more precise predictions.