The impact of incomplete landslide inventories on susceptibility mapping: assessment and solutions

Date

2024-6

Author

Kantekin, Sertaç

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

119
views

149
downloads

Cite This

This study focuses on measuring the impact of inventory completeness and incomplete inventory on landslide susceptibility mapping and providing solutions to reduce this impact. In this context, landslide inventory data from the Melen region, which is considered to be well-established and complete, was used. To measure the impact of incomplete inventory on landslide susceptibility mapping, from this inventory, landslide susceptibility maps were generated using Random Forest, Decision Trees, and Logistic Regression methods. Among the generated maps, the method with the highest AUC (Area Under Curve) score was Decision Trees, and the landslide susceptibility map produced by this method was designated as the benchmark susceptibility map representing the region's landslide susceptibility most accurately. Successively, new landslide susceptibility maps were generated by reducing the inventory completeness by 10%, and the differences between the sensitivity classes of the generated maps and the benchmark map were measured. Root Mean Square Error (RMSE), confusion matrices, Difference maps, F1 scores, and Area Under Curve (AUC) scores were used for measurement. As a result of the measurements, it was found that as the inventory incompleteness increases 10%, the similarity of the maps generated from incomplete inventories to the benchmark map decreases. Additionally, it was concluded that AUC (Area Under Curve) scores were not a suitable metric for this measurement. Two methods whose names were Synthetic Seed Cell Generation (SSCG) and Area Based Synthetic Seed Cell Generation (ASSCG) were applied to reduce the impact of inventory incompleteness on landslide susceptibility mapping. The maps obtained from these methods were also compared with the benchmark map, and the maps generated for each inventory using the ASSCG showed a greater similarity to the benchmark map than maps obtained from incomplete inventories and maps obtained from SSCG.

Subject Keywords

Landslides, Landslide Susceptibility Mapping, Landslide Inventory, Incomplete Inventory

URI

https://hdl.handle.net/11511/110078

Collections

Graduate School of Natural and Applied Sciences, Thesis