EGU2020-6749, updated on 10 Jan 2022
https://doi.org/10.5194/egusphere-egu2020-6749
EGU General Assembly 2020
© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

Dormant and Active Landslides Classification Using Machine Learning Algorithms Coupled With Geological Field Inspection: Pohang Case Study

Omar F. Althuwaynee1, In-Tak Hwang1, Hyuck-jin Park1, Swang-Wan Kim1, and Ali Aydda2
Omar F. Althuwaynee et al.
  • 1Department of Energy and Mineral Resources Engineering, Sejong University, Seoul, Republic of Korea
  • 2Department of Geology, Faculty of Sciences, Ibn Zohr University, Agadir, Morocco

In 1998, intense rainfall events hit the Pohang state, south west of Korea, which results in highest number of landslides registered in this area (generally the area has a relatively short history of landslide inventorying). The current inventory was digitized using Aerial photographs (lack of photogeological stereoscopic analysis of the aerial images) and coupled with basic field verification (due to limit funding available). Leaving the applied susceptibility maps models performed, using this inventory, with high degree of uncertainty.  Currently a research initiative carried to audit the landslide inventory using freely available aerial photographs and the time tuning function in Google earth for aerial archives. We notice some slopes area covered with deformed forest types that is similar in texture to that seen in digitized locations of landslides inventory. Due to long retune period of similar rainfall event, and with an assumption that the available landslides inventory might not complete. A certain hypothesis of additional investigation including field work to audit the landslides incidents is highly needed. In the current research, we assumed that, some dormant slopes caused by the 1998 event can be reactivated with the current extreme (uncontrolled) uses of slopes by human activities (constructions of real estate’s projects). To that end, a methodology of three main stages were proposed.

Stage one; Dormant susceptibility map (DSM) coupled with landslide susceptibility map will be produced. Machine learning supervised classification of eXtreme Gradient Boosting algorithms and Ensemble Random Forest, that run on tree-based classification assumption considering only active and dormant landslides as well as stable ground. Stage two; field work needs to be designed by geological and geotechnical experts to collect the doubtful locations by guidance of DSM and consider the new locations as dormant inventory. However, the areas of low dormant susceptibility (or mutual zones with Landslide susceptibility) will be recommended for advanced filed work and soil sampling test to complete the landslides identification of such highly urbanized area. Stage three; knowing the contour depths of diluvial and alluvial deposits can be useful for extracting areas that are more prone to landslides. Especially in the case of a rigid bedrock beneath the diluvial crust. Therefore, reconstructing the Quaternary formation thickness using boreholes repository and then represent the entire study area using CoKriging surface interpolation technique with elevation model. The current research results will provide us a better understanding of landcover stability conditions and their spatial prediction features.

 
hjpark@sejong.edu
omar.faisel@gmail.com

How to cite: Althuwaynee, O. F., Hwang, I.-T., Park, H., Kim, S.-W., and Aydda, A.: Dormant and Active Landslides Classification Using Machine Learning Algorithms Coupled With Geological Field Inspection: Pohang Case Study, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-6749, https://doi.org/10.5194/egusphere-egu2020-6749, 2020.

Displays

Display file