Comparative Study on the Effectiveness of Landslide Susceptibility Assessment Based on Different Evaluation Units and Models

ABSTRACT To investigate the evaluation performance of different models across various evaluation units, 174 landslide samples were selected from Xide County, Sichuan Province, China, as the study area, considering 12 conditioning factors such as aspect, slope and elevation. Using software tools such as ArcGIS and SPSS, the landslide susceptibility in the study area was assessed across different units (12.5 and 30 m grid units and slope units). Four models were employed for this evaluation: the information value model (IV), the logistic regression model (LR), the information value–logistic regression coupled model (IV‐LR) and the decision tree model (DT). The evaluation accuracy of various models across different evaluation units was analysed using rationality testing and ROC curves. The results indicate that, within the same evaluation model, the landslide susceptibility assessment accuracy for the 12.5 m grid unit surpasses that of the other two evaluation units, with an average AUC value of 0.849. Under the same evaluation unit, the IV‐LR coupled model consistently demonstrated strong performance across all units, achieving the highest AUC value of 0.881 with the 12.5 m grid unit.