SURFACE GEOSTROPHIC CURRENT PREDICTION IN INDONESIA WITH ARTIFICIAL NEURAL NETWORKS

• Main Author: Nadhif Taher Ahmad, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/71169
• Institution: Institut Teknologi Bandung

This research conducts daily geostrophic current speed predictions to determine the performance of the machine learning method used. The prediction is made for next five days with a training data range from January 1, 2013 to December 31, 2018 within Indonesian region and verification data is within January 1, 2019 to August 2, 2021. The data used is from satellite observations. The method used is artificial neural networks with three scenarios used as input, namely absolute dynamics topography (ADT), U (previous day zonal speed) and V (previous day meridional speed), as well as ADT+UV. The ADT+UV scenario is the best scenario because it has a better correlation value and (normalized root mean square error) NRMSE than the other scenarios. Performance metrics will deteriorate as the predicted day becomes further away. The best performance was obtained for the next first day (H+1) with a zonal speed correlation value of 0.973 and meridional speed correlation value of 0.964 Observing the spatial patterns, geostrophic currents in the Sulawesi Sea move westward throughout the year and in the Java Sea move westward in January. In addition, Indonesia has a large geostrophic speed of 0.280 m/s with January being the fastest at 0.233 m/s.