EGU2020-4090
https://doi.org/10.5194/egusphere-egu2020-4090
EGU General Assembly 2020
© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

DTM-based landslide detection using deep learning: A case study in Hong Kong

Haojie Wang and Limin Zhang
Haojie Wang and Limin Zhang
  • The Hong Kong University of Science and Technology, School of Engineering, Civil and Environmental Engineering, Hong Kong (h.wang@connect.ust.hk)

Landslide detection is an essential component of landslide risk assessment and hazard mitigation. It can be used to produce landslide inventories which are considered as one of the fundamental auxiliary data for regional landslide susceptibility analysis. In order to achieve high landslide interpretation accuracy, visual interpretation is frequently used, but suffers in time efficiency and labour demand. Hence, an automatic landslide detection method utilizing deep learning techniques is implemented in this work to conduct high-accuracy and fast landslide interpretation. As the ground characteristics and terrain features can precisely capture the three-dimensional space form of landslides, high-resolution digital terrain model (DTM) is taken as the data source for landslide detection. A case study in Hong Kong, China is conducted to validate the applicability of deep learning techniques in landslide detection. The case study takes multiple data layers derived from the DTM (e.g., elevation, slope gradient, aspect, etc.) and a local landslide inventory named enhanced natural terrain landslide inventory (ENTLI) as its data sources, and integrates them into a database for learning. Then, a deep learning technique (e.g., convolutional neural network) is used to train models on the database and perform landslide detection. Results of the case study show great performance and capacity of the applied deep learning techniques, which provides valuable references for advancing landslide detection.

How to cite: Wang, H. and Zhang, L.: DTM-based landslide detection using deep learning: A case study in Hong Kong , EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-4090, https://doi.org/10.5194/egusphere-egu2020-4090, 2020.