Machine Learning-Based Landslide Susceptibility Mapping and Environmental Controls in Hanang District, Northern Tanzania

Abstract Landslides are among the most destructive geohazards in mountainous regions, causing substantial socio-economic and environmental challenges. In Tanzania, the 2023 landslide disaster in Hanang district highlighted the need for reliable spatial tools to support landslide risk assessment and disaster preparedness. This study employed a Random Forest (RF) machine learning model, integrated with remote sensing and Geographic Information Systems (GIS), to map landslide susceptibility in the study area. A total of 859 inventory points, comprising 429 observed landslide locations and 430 non-landslide locations, were used to develop and validate the model. A comprehensive set of ten (10) environmental conditioning factors was derived from the satellite imagery and geospatial datasets, including annual rainfall, rainfall event intensity, elevation, slope, aspect, curvature, normalized difference vegetation index (NDVI), topographic wetness index (TWI), distance to rivers, and land use/land cover features. The RF model achieved an overall accuracy of 93.5% and a Kappa coefficient of 0.869, demonstrating outstanding predictive performance. Variable importance analysis identified NDVI as the dominant predictor of landslides. The susceptibility map indicated that approximately 43.92% of the district falls within the high and very high susceptibility classes, representing the area’s most vulnerable to future landslides. Spatial validation further demonstrated strong model reliability, with 93.71% of the observed landslide inventory located within the high and very high susceptibility zones. These findings demonstrate that the RF model effectively predicts landslide susceptibility and provides valuable information for disaster risk reduction, land-use planning, and climate adaptation strategies.