An approach for multi-hazard susceptibility assessment for landslides, earthquakes and floods

Production of precise and up-to-date susceptibility maps at regional level is essential for mitigating disasters, selecting new sites for settlements and construction, and planning in areas prone to various natural hazards. This thesis introduced a novel approach to multi-hazard susceptibility assessment (MHSA) for evaluating landslide, flood, and earthquake risks, combining expert opinion with supervised machine learning (ML) techniques. The methodology was tested in five basins within Elazig and three basins in Adiyaman Provinces, Türkiye. The susceptibility maps were produced at basin scale since various environmental characteristics affecting the hazard conditioning factors are relatively coherent within them. Regarding landslide susceptibility mapping (LSM), the random forest (RF) ensemble machine learning algorithm, was utilized. For flood susceptibility mapping (FSM), a modified analytical hierarchical process (m-AHP) method was employed using factor scores provided by experts for each site. Seismic hazard assessment relied on ground motion parameters, specifically Arias intensity values, as they are considered to be effective especially for landslides. These individual assessments were then synthesized using a Mamdani fuzzy inference system (FIS), incorporating expert-defined membership functions. The thesis findings indicated high overall accuracies (over 90%) can be achieved with the random forest model for the LSM. The Mamdani fuzzy algorithm is recommended for the MHSA, as it can be adapted to different regions with its intuitive membership functions. While the thesis provided a robust framework for multi-hazard susceptibility assessment at the regional scale, fine-tuning of the algorithms in different geographical areas may require further expert input.