APPLICATION OF DEEP LEARNING ALGORITHM FOR PREDICTING THE TSUNAMI HEIGHT AND ARRIVAL TIME ON BALI ISLAND

• Main Author: Iqbal Ikromi, Akhmad
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/73612
• Institution: Institut Teknologi Bandung

<p align="justify"> Indonesia, located in the Ring of Fire zone, is prone to earthquakes and tsunamis. Bali Island is an internationally renowned tourist destination known for its beautiful beaches. Due to its location on the southern side of Indonesia and its direct proximity to the Java Megathrust subduction zone, as well as its population concentrations along coastal areas, Bali is highly vulnerable. One effective method to mitigate the impact of tsunamis is by implementing a robust early warning system that provides timely information about the tsunami's height and arrival time. This research utilizes deep learning algorithms to predict tsunami parameters. To simulate 21 major scenarios (ranging from 7.0 to 9.0 with 0.1 Mw intervals), we employed the linear shallow water equation and solved it using the Delft3D program. Observations were conducted at 6 beach tourism destinations in the south of Bali. The database was divided into three parts for training, testing, and validation purposes. The deep learning programming was implemented using Python with the TensorFlow library. The input layer consisted of 7 neurons representing variable locations and magnitudes, followed by 2 hidden layers with 50 and 100 neurons, respectively. The output layer consisted of 120 neurons, representing the height of the tsunami every minute for a duration of 2 hours. The correlation values obtained for the maximum tsunami height and arrival time were 0.985 and 0.919, respectively. Deep learning algorithms reduced the size of the database by 33.3%. Furthermore, compared to numerical simulation, deep learning algorithms demonstrated a calculation speed that was 15,000 times faster, while maintaining a correlation coefficient above 0.90.