Landslide susceptibility assessment using uncertain decision tree model in loess areas

Because of the complexity of the causative factors and the uncertainty in their measurement, it is generally difficult to analyze them quantitatively and to predict the probability of landslide occurrence. A major issue with landslide susceptibility analysis models based on decision tree algorithm is the difficulty in quantifying triggering factors (precipitation). To address this issue, a new method based on an uncertain decision tree algorithm (DTU) is proposed to assess landslide susceptibility model. Thematic maps representing various factors related to landslide activity were generated using GIS technology. Areas susceptible to landslides were analyzed and mapped in the city district by the ID3, C4.5 and DTU algorithms using the same landslide-occurrence factors. For the quantitative assessment of landslide susceptibility, the accuracy of the area under the curve (AUC) in the ID3 and C4.5 algorithms was 83.74 and 85.89%, respectively; the accuracy of the AUC using the DTU algorithm was 89.25%. The prediction accuracy of the DTU model for the landslide susceptibility zone map is greater than the accuracy of the ID3 and C4.5 algorithms. Thus, the DTU algorithm can be utilized capably for landslide susceptibility analysis and has the potential to be widely applied in the prediction spatial events.