Landslide Susceptibility and Slope Instability Modeling Using GIS Based ANN and MT-INSAR Techniques

No Thumbnail Available



Journal Title

Journal ISSN

Volume Title


Addis Ababa University


Landslides are common known natural hazards that lead to infrastructure destruction (road failure), property damage, and human loss. Current study area located in southwestern Ethiopian highlands near rift margin and proned to landslide. Spatial and temporal landslide susceptibility studies are important for predicting, preventing, and mitigating landslide-derived threats. This study was aimed to analysis role of landslide causative factors (weight determination), predict spatial landslide susceptibility and model slope instability based on spatial coherence preliminary result. GIS based ANN Landslide susceptibility was performed through the identification and analysis of hill slope instability factors and landslide inventory. A total of 195 polygons and twelve thematic rater pixels were intersected from which 1482 training and 513 testing pixels were extracted to train the model. Simultaneously, the ANN model was trained to achieve an optimum standard error between the predicted and observed dataset by adjusting weights and NN structures. The generalized weight result revealed the factors effect on the landslide occurrence positively, negatively and neutrally. The LSS result showed that 213.6 (86.7%), 3 (1.2%), and 29.4 (11.9%) areas fall into low, medium, and highly susceptible zones, respectively. in other side, cumulative displacement was modeled based on the preliminary interferometric coherence result. freely available 20 sentinel-1A SLC imageries from 2017 to 2021 were processed for this purpose. Spatial coherence test revealed that it is impossible to monitor and model slope instability (cumulative displacement) entire the landslide prone area. Attempt was made to model cumulative displacement at partially coherent landslide susceptible sample area. The result showed that there was cumulative displacement rate ranging from -6 mm/yr to +6 mm/yr.ROC and AUC are important tools to test training and predicting capability of machine learning model the model achieved training and testing performance of 94% and 88% respectively. This indicated that the model had successfully learned the existing problem and predicted it for the future. Therefore, the susceptibility result can be used to plan future development activities. Spatial coherence revealed, there is need of using high resolution SAR datasets to monitor slope instability entire the study area.



Landslide, Artificial Neural Network, Landslide Inventory, Spatial Coherence, Synthetic Aperture Radar