Integration of GIS and Environment-Based Machine Learning Variables for Flood and Landslide Analysis Sirimau Sub-District, Ambon City, Indonesia

Abstract

Sirimau District often experiences floods and landslides during the rainy season. This study uses environmental variables and the coordinates of flood and landslide locations for MaxEnt modeling. The results show that elevation and land cover are the most influential factors for floods (70.3% and 22.9%, respectively) and landslides (80.9% and 10.3%), consistent with hydrology and physical geography theories. The flood and landslide vulnerability levels are divided into three classes, with low and moderate risk areas dominating, while high-risk areas require special attention for stricter management. Model validation with high Area Under Curve (AUC) values (0.973 for floods and 0.845 for landslides) ensures prediction reliability, which can serve as a basis for adaptive spatial data-based mitigation policy making. Policy recommendations include strengthening early warning systems, spatial planning based on risk zoning, and community capacity building, which are expected to reduce social and economic impacts from disasters in this area sustainably.