Assessment of the effect of point and polygon-based landslide inventory on landslide susceptibility mapping

This study investigated the impact of point and polygon-based landslide inventory on the landslide susceptibility mapping in the Kysuca river basin, Slovakia. In this regard, logistic regression (LR), random forest (RF), and extreme gradient boosting (XGBoost) models were used. For each model, polygon-based and point-based landslide susceptibility map were constructed and compared. The results showed that polygon-based landslide susceptibility map outperformed the point-based landslide susceptibility map in both accuracy and consistency. In particular, RF and XGBoost performed significantly higher for polygon-based landslide susceptibility map compared to that of point-based landslide susceptibility map. For the point-based approach, the success rate/prediction rates were 73.84/70.69%, 92.87/80.97%, 96.29/85.59%, while those values for the polygon-based approach were 76.66/75.29%, 97.82/95.14%, 98.56/96.12%, respectively for LR, RF and XGBoost models. Both the point and polygon-based inventories provided satisfactory accuracy but using the polygon-based approach is recommended if data and computing sources are available.