An application of deep learning for lightning prediction in East Coast Malaysia

Abstract

This paper presents the application of deep learning (DL) approach namely Feed-Forward Neural Networks (FFNN) in predicting the location of lightning occurrences within 100 km radius from Universiti Malaysia Pahang Al-Sultan Abdullah (UMPSA) Pekan, Pahang Malaysia. The recorded data were obtained from Malaysia Meteorology Department (MET Malaysia), where the inputs of the DL are the intensity of the lightning in kilo Ampere, direction in degrees, distance and major axis that measures in km, while the output is the latitude and longitude of the lightning occurrences. The data are divided into training, validation and testing to measure the performance of the developed DL model. The findings of the study demonstrated the promising results of FFNN in terms of obtaining the minimum error which significantly increasing the accuracy of the predictions. To show the effectiveness of FFNN, the comparison study has been conducted with Long Short-Term Memory (LSTM) networks. From the simulation, it can be seen that FFNN can be used as an effective tool for predicting the location of lightning occurred better than the LSTM.