Detecting Post-Earthquake and Tsunami Mangrove Dynamics in the Nias Islands Using a Machine Learning Approach

Abstract

This study presents an approach to detecting and analyzing the dynamics of mangrove vegetation cover in the Nias Islands and surrounding areas following the 2004 and 2005 earthquakes and tsunamis. A predictive model was developed using a machine learning approach with a decision tree algorithm on the 2025 dataset. This study utilized several indices derived from Landsat 5 TM and 9 OLI, along with the derivation of socio-geo-biophysical data such as elevation, slope, substrate, distance from rivers, and coastlines.  The analysis of land cover change trajectories was focused on the mangrove density (moderate and dense mangrove density, non-mangrove vegetation, bare land, settlements, paddy fields, water bodies, and non-land cover categories such as clouds. The study found that the developed decision tree model can accurately detect mangrove vegetation and other land cover types.  The model evaluation results showed that the F1 score for all classes exceeded 90%. The overall accuracy was 95.9%. In general, the study found that mangrove loss in the Nias Islands and surrounding areas reached 6.9% between 2003 and 2025.  This study confirms that the combination of optical imagery and socio-geo-biophysics data yields good performance in detecting and mapping the dynamics of changes in mangrove vegetation cover.