Landslide susceptibility mapping in southwest Sweden using an automated deep neural support vector regression model

¹ JohanLundberg AB, Uppsala, Sweden
² KTH Royal Institute of Technology, Stockholm, Sweden

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Landslides are a persistent geohazard in Sweden, especially in areas dominated by soft, clay-rich soils. This study develops a high-resolution landslide susceptibility map using a novel automated deep neural support vector regression (ADNSVR) model. Seventeen conditioning factors were included, grouped into remote-sensing-, geo-, land-, and distance-based parameters. Geo-based factors included soil depth, soil type, and soil type, highlighting the importance of sensitive clays. The model was trained and validated with documented Swedish landslides and showed strong predictive performance. Sensitivity analysis identified soil type, soil depth, slope, and soil type as the most influential controls. Glacial and postglacial clays, covering about 32% of the area, were the most significant sub-types. These results support regional landslide susceptibility assessment and risk-informed land-use planning.